---

tags: user support coffee break
title: archive 2

---
# Archive of Weekly CSC user coffee break Q&As (vol2)

**Old archive: https://hackmd.io/9QsIbJ03T1SaNDV_xbT8Dw?view**

**Contents of this documents and quicklinks**
[TOC]

## Recordings, slides etc material from short talks

7.9. Puhti web interface ([recording](https://video.csc.fi/media/t/0_j18nkvvh))
14.9. LUMI ([slides](https://a3s.fi/mlehtiva-2004306-pub/LUST_CSC_coffee_break_20220914.pdf), [recording](https://video.csc.fi/media/t/0_bz8m2kbn))
21.9. CSC Notebooks ([slides](https://a3s.fi/mlehtiva-2001659-pub/CSC_Notebooks_Sept2022.pdf), [recording](https://video.csc.fi/media/t/0_uhgggpml))
28.9. Puhti OS update ([slides](https://a3s.fi/swift/v1/AUTH_53f5b0ae8e724b439a4cd16d1237015f/mlehtiva-2001659-pub/RHEL8-flash.pdf))
5.10. Tykky ([slides](https://a3s.fi/swift/v1/AUTH_53f5b0ae8e724b439a4cd16d1237015f/mlehtiva-2001659-pub/tykky.pdf), [recording](https://video.csc.fi/media/t/0_3qy1o8uv))
12.10. SD services (slides, [recording](https://video.csc.fi/media/t/0_yde0f1iq) )
19.10. No short talk (autumn holidays)
26.10. Machine Learning guide (no slides, all information in [docs](https://docs.csc.fi/support/tutorials/ml-guide/), [recording](https://video.csc.fi/media/t/0_na3upt7e))
2.11. CSC cloud services Pouta and Rahti ([slides](https://cloud-solutions.a3s.fi/researchsupportweeklyzooms/index.html), [recording](https://video.csc.fi/media/t/0_auvrto32))
9.11. Workflows and Hyperqueue ([recording](https://video.csc.fi/media/t/0_1u744x4n))
16.11. CSC Training materials ([recording](https://video.csc.fi/media/t/0_k65fmzx9))
23.11. RStudio/R in different platforms ([slides](https://a3s.fi/heli-slides/User_zoom_Rintro_23112022.pdf), [recording](https://video.csc.fi/media/t/0_t0pp68xr))
30.11. Introduction to the Language Bank of Finland (Kielipankki) ([slides_part1](https://a3s.fi/swift/v1/AUTH_53f5b0ae8e724b439a4cd16d1237015f/mlehtiva-2001659-pub/Kielipankki_Martin_UserSupportCoffee.pdf), [slides_part2](https://a3s.fi/swift/v1/AUTH_53f5b0ae8e724b439a4cd16d1237015f/mlehtiva-2001659-pub/Kielipankki_Sam_User_support_coffee_2022-11-30.pdf), [recording](https://video.csc.fi/media/t/0_u6z4grq2) )
7.12. [MyCSC](https://my.csc.fi) ([recording](https://video.csc.fi/media/t/0_33ipi59b))
14.12. [mini crash course in HPC](https://a3s.fi/swift/v1/AUTH_53f5b0ae8e724b439a4cd16d1237015f/mlehtiva-2001659-pub/mini-crash-course-HPC.pdf)
11.1. Upcoming CSC Computing Environment course ([Self-learning course page](https://ssl.eventilla.com/csccompenvselflearn))
18.1.Data retention policy updates ([Slides](https://a3s.fi/mlehtiva-2004306-pub/support-coffee-data-retention-policies-scratch-disk-18012023.pdf), [recording](https://video.csc.fi/media/t/0_a188372q), [data clean up in docs.csc.fi](https://docs.csc.fi/support/tutorials/clean-up-data/))
25.1.: No short talk, just Q&A session
1.2.: No short talk, just Q&A session
8.2.: Mini crash course in HPC 2 ([Slides](https://a3s.fi/mlehtiva-2001659-pub/mini-crash-course-batch-job.pdf))
15.2.: No short talk
22.2.: SD Desktop ([recording](https://video.csc.fi/media/t/0_83cphk52)
1.3.: LUMI ([Slides](https://a3s.fi/weekly-user-zoom/CSC_coffee_break_20230301.pptx)[recording](https://video.csc.fi/media/t/0_tnh1lrun))
15.3.: R/python-based web tools at CSC for reproducible research and teaching ([Slides](https://a3s.fi/csc-training/weekly_customer_csc.html))
21.3.: No short talk
29.3.: SSH keys and login safety ([recording](https://video.csc.fi/media/t/0_co757hyk))
5.4.: Mini course as a first step to CSCs services ([mini course](https://e-learn.csc.fi/course/view.php?id=75))
12.4.: Help us help you! or: How to write a good support request ([recording](https://video.csc.fi/media/t/0_6z98kv08))
3.5.: Monitoring batch jobs on supercomputer
10.5.: Short talk: How to use the Julia programming language on CSC’s Puhti and Mahti clusters
https://docs.csc.fi/apps/julia/
https://docs.csc.fi/support/tutorials/julia/
17.5.: No short talk
24.5.: [GPU programming course](https://enccs.se/events/2023-06-gpu-programming/)
2024
3.4.: Machine learning on supercomputers, Part 3: Multi-GPU and multi-node jobs** [Slides PDF](https://a3s.fi/mats/multi-gpu-multi-node.pdf)

## 2024-04-03 session

**Short talk: Machine learning on supercomputers, Part 3: Multi-GPU and multi-node jobs** [Slides PDF](https://a3s.fi/mats/multi-gpu-multi-node.pdf)

### Zoom breakout rooms and topics

Room 1 - Beginner/getting started-questions (Laxmana)
Room 2 - Machine learning + Q234 + Q235 (Mats)
Room 3 - R (Panu & Heli)

### Questions

- **Q231** I would like to use CSC Notebooks as a collaboration platform. How should I organise the workspace? I need multiple users to be able to work through the same Jupyter notebook with the same data, but in Docs it said that you're not supposed to put the actual notebook files in the shared folder. So I'm wondering what goes where. - Ani
- Notebooks specialists are not right now available, please type a message to service desk ([e-mail]), they will take this question there. Needs a bit of pondering. You can reserve a Zoom meeting with them and/or use these Wednesday meetings!

- **Q232** Probably a newbie/basic question, but I want to run some R analysis in a non-interactive session and store the results to disk. Some basic scripts that I don't want to run on a laptop. -Panu
- Discussed in room 3

- **Q233** My home folder is getting full. I am not sure what is causing that and which files are safe to remove from there to free space. - Namel
- You can use the [LUE tool](https://docs.csc.fi/support/tutorials/lue/) to find out what is taking up the space in your home directory. Typical files include various cache and session files that can usually be removed. You may also have some installations under $HOME/.local if you have used e.g. `pip install --user`. These should not be removed (unless you are sure they are no longer needed), but you should consider using /projappl for them instead. You may also find the FAQ entry ["Disk quota exceeded"](https://docs.csc.fi/support/faq/disk-quota-exceeded/) helpful.

- **Q234** Trouble shoot: I am loading pytorch with `module load pytorch`. Does it have pytorch_lightning by default?
I get `ModuleNotFoundError: No module named 'pytorch_lightning'` error when trying to load the pytorch lightning. Then, when I try to install with pip in the batchscript, it give this error `OSError: [Errno 28] No space left on device`. I have enough space left on workspaces.- Gantugs
- The PyTorch module has `pytorch_lightning` available (I just double checked). As for the other issues, we can take a look in the breakout room.
- => **Go to breakout room 2!**

- **Q235** How to connect to LUMI? Using one GPU works automatically but How could I modify my code to use 4 GPU? any examples ? [Mohamed]
- => **Go to breakout room 2!**

- **Q236** CST hasn't installed on my project yet. Whom I need to contact to for installation? Quoc Duy Nguyen.
- A: Could you specify a bit? Which software are we talking about?
- Yes, sorry I can't talk. I mean CST Microwave Studio.
- Anyway, thank you for your time. I will ask my manager first.
- Esko Järvinen might know something about this.
- => email to: [e-mail] (this reaches all our specialists, so Esko too :) )
- Thank you.

## 2024-03-27 session

Note: [LUMI coffee break](https://www.lumi-supercomputer.eu/events/usercoffeebreaks/) at the same time today!

**Short talk:** CSC courses and training material
- Upcoming courses
- list below and https://www.csc.fi/training
- [LUMI trainings](https://www.lumi-supercomputer.eu/events/)
- [Aalto SciComp trainings](https://scicomp.aalto.fi/training/)
- [HPC trainings in Europe](https://hpc-portal.eu/visualise_new_content) collection
- [ENCCS trainings](https://enccs.se/events/)
- Get the info, subscribe to Training newsletter https://www.csc.fi/newsletter (once a month summary of upcoming trainings)
- Ongoing online self-learning courses:
- [Elements of Supercomputing](https://ssl.eventilla.com/event/mlOk6)
- [Research data management](https://ssl.eventilla.com/event/v8B6B)
- [CSC Computing Environment](https://ssl.eventilla.com/csccompenvselflearn)
- Topical courses, like [Practical machine learning](https://e-learn.csc.fi/enrol/index.php?id=14), [Single-cell RNA-seq data analysis with Chipster](https://e-learn.csc.fi/course/view.php?id=17)...
- Available training material
- [Training material collection](https://docs.csc.fi/support/training-material/), [Tutorials](https://docs.csc.fi/support/tutorials/#bioinformatics) and [FAQs](https://docs.csc.fi/support/faq/) in Docs.csc.fi
- [CSC Notebooks = CSC Noppe](https://notebooks.rahtiapp.fi/)
- [eLena](https://e-learn.csc.fi) = CSC Moodle environment
- LOTS available, google or ask us!

### Zoom breakout rooms and topics

Room 1 - Beginner/getting started-questions, Mats Q224
Room 2 - Q229 and Q223 Allas / Jupyter, Kimmo
Room 3 - Matlab or Julia / Jaan
Room 4 - Installations / Samantha

### Questions

- **Q223** Hi, I am using interactive Jupyter on Puhti, and I'd like to upload a file from Scratch to Allas, using a command directly from jupyter.
'!module load allas' works, Allas tools loaded.
'!allas-conf' doesn't, the output:'/bin/bash: allas-conf: command not found'
What am I doing wrong? Thank you! / Taras
- Kimmo, in room 2

- **Q224** If my Masters starts in May, I should not make the project yet because the time is only 6 months for the project? And if it's free, why do I need the project manager (terms in use)?
- This student project is a bit new feature, so we would instruct to make the project now when you need it, and then after 5 months just email servicedesk and ask for more time.

- **Q226** Hi, I have problem in Lumi. I have two project number but I could only access one. + Whenever i want to access the server, it requirees passwords. / Choo
- To LUMI call: [LUMI coffee break](https://www.lumi-supercomputer.eu/events/usercoffeebreaks/)

- **Q227** About installing tools in Puhti (Python & Conda env) / Kisun Pokharel and Melak Weldenegodguad
- Best practices for installing own python tools on Puhti
- check if readymade module available (recommended especially for R)
- own conda installation using Tykky
- start with one environment file, without versions (unless you need to pin some packages to specific version)
- if you later need to add more files use the `update` functionality
- check which versions of your packages were installed to your environment and store them (helps with if you need to recreate (more or less) exact environment later, eg to adress reviewer comments)
- environment files can be tracked for example using git
- Tykky environment for Python2 and separate environment for Python3, keep separate
- strong recommendation to move away from Python2 (if possible)

- **Q229** I've been a bit careless with storing data on the scratch. I seem to have discrepancy between what csc-workspaces report as opposed to what du gives, why is this? I could also briefly chat about data storage best practices in general. / VT
- Kimmo, in room 2

- **Q230** Jaan Tollander de Balsch, see CSC #690787.
- Lets wait we get a breakout room for MATLAB (and Julia)
- Room 3
- Problem resolved.

## 2024-03-20 session

**Short talk:** No short talk

**Advertisements:**

- [Nextflow hackathon](https://ssl.eventilla.com/event/9kgb0) on 11.4.2024, registration for waitlist open.
- [Using CSC Computing Environment, Part 1: Basics](https://ssl.eventilla.com/part1april24) 24-25.4.2024, registration open.
- [Data Analysis with R](https://ssl.eventilla.com/event/v8Kv9) on 6.-7.5.2024, registration open.
- [Using CSC Computing Environment, Part 2: Next steps](https://ssl.eventilla.com/part2may24) 15-16.5.2024, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6.-4.7.2024, registration open. Early Bird Fee deadline approaching. Last chances to get discounted price!

### Zoom breakout rooms and topics

Room 1 - Getting started with SD (Kimmo)
Room 2 - Company use accounts / machine learning (Mats)
Room 3 - Beginner/getting started-questions (Cristian)
Room 4 -
Room 5 -
Room 6 -

## 2024-03-13 session

**Short talk:** [CSC Cloud Services for users: A couple of use cases](https://a3s.fi/media/csc_cloud_services_use_cases_13_03_24.pdf)

**Advertisements:**

- [Nextflow hackathon](https://ssl.eventilla.com/event/9kgb0) on 11.4.2024, registration for waitlist open.
- [Using CSC Computing Environment, Part 1: Basics](https://ssl.eventilla.com/part1april24) 24-25.4.2024, registration open.
- [Data Analysis with R](https://ssl.eventilla.com/event/v8Kv9) on 6.-7.5.2024, registration open.
- [Using CSC Computing Environment, Part 2: Next steps](https://ssl.eventilla.com/part2may24) 15-16.5.2024, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6.-4.7.2024, registration open.

### Zoom breakout rooms and topics

Room 1 - Beginner/getting started-questions (Mats)
Room 2 - Cloud services / Joona
Room 3 -
Room 4 -
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q220** How to stop a runnng job in puhti/Andry
- Answer: command `scancel <JOB_ID>` where you replace `<JOB_ID>` with the job ID (number) of your job. (Command `squeue --me` show)

- **Q221** I run snakemake pipeline with config file setting in Tykky as recommended. However I always get the same error "/var/spool/slurmd/job20782041/slurm_script: line 3: /CSC_CONTAINER/miniconda/envs/env1/bin/python3.10: No such file or directory". I tried to run the individual task independently (outside snakemake), and it was perfectly fine. Did I miss some setting? / Pande
- Waiting to see if a specialist can join the session. If not, please send an email to [e-mail]
- Answer: You can manually patch the snakemake executable to make it work.

so create a file e.g `post.sh` with this contents:

        sed -i 's@#!.*@#!/Path/to/python/bin/python<ver>@g' $env_root/bin/snakemake

Where the path in the sed command would probably be `/projappl/project_200xxxx/env/bin/python3.11` (check that it actually exists)

Then update the installation:

        conda-containerize update <path to installation> --post-install post.sh

- The error apparently related to the snakemake version loaded. Because the snakefile will refer to the package that was installed using tykky, the compatible snakemake version is 7.17.1. Other version seems to failed with different error. See more: https://docs.csc.fi/support/tutorials/snakemake-puhti/

- **Q222** I've been granted access to the LUMI CSC service for my project. However, the project hasn't appeared in my LUMI account yet. How long does it usually take for it to appear in my account? / Omid
- Typically a couple of hours I think. Did you also approve the LUMI terms and conditions in MyCSC? In the project page, under "Services" there should be a "LUMI" button that you can click open to approve it.
- The service status is as 'Enabled' in my account!
- Has it been a long time? Does it show up on LUMI if you run the command `lumi-allocations`?
- It has been enabled this morning. Yes it does!
- Then it should work on the LUMI side. Is there something that doesn't work? If you have an old ssh session open the groups of your users might not show correctly.
- I can't see it in my Home Directory.
- Might be best to contact LUMI support directly. https://lumi-supercomputer.eu/user-support/need-help/account/
- If you're referring to the LUMI web user interface, try clicking "Help" in the menu and then "Restart Web Server"
## 2024-03-06 session

**Short talk:** -

**Advertisements:**

- [Biowebinar: Recent updates for bioscience users at CSC](https://ssl.eventilla.com/event/WLObB) on 12.3.2024 @ 14:00, registration open.
- [Nextflow hackathon](https://ssl.eventilla.com/event/9kgb0) on 11.4.2024, registration open.
- [Using CSC Computing Environment, Part 1: Basics](https://ssl.eventilla.com/part1april24) 24-25.4.2024, registration open.
- [Data Analysis with R](https://ssl.eventilla.com/event/v8Kv9) on 6.-7.5.2024, registration open.
- [Using CSC Computing Environment, Part 2: Next steps](https://ssl.eventilla.com/part2may24) 15-16.5.2024, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6.-4.7.2024, registration open.

### Zoom breakout rooms and topics

Room 1 - Beginner/getting started-questions (Kimmo)
Room 2 - Rahti & cPouta (Mehrdad & Jemal)
Room 3 -
Room 4 -
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q218** Update on Rahti v2/ What is the realistic date for the release? + Mehrdad

- **Q219** Is cPouta a right service to run an Apache web servie on it? / Mehrdad
- It depends on what you are going to serve from the web server, but yes. It is an easy way to run a web server, small or big, accesible to everyone in the internet or only a selected range of ips, ... What is the concern here? - Alvaro
- Great. We will run our Django project on it and will come back to you after that if we encounter any issues.

## 2024-02-28 session

**Short talk:** Using MATLAB on Puhti

**Advertisements:**

- [Introduction to Geospatial Python](https://ssl.eventilla.com/geospatial_python) on 4.-6.3, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6.-4.7., registration open.
- [Biowebinar: Recent updates for bioscience users at CSC](https://ssl.eventilla.com/event/WLObB) on 12.3.2024 @ 14:00
- [Nextflow hackathon](https://ssl.eventilla.com/event/9kgb0) on 11.4.2024

### Zoom breakout rooms and topics

Room 1 - Beginner/getting started-questions
Room 2 - Matlab (Jaan)
Room 3 - Puhti | Problem with package installation (Mats)
Room 4 -
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q210** In cPouta instance, how to restrict access to/from only University of helsinki? / GT
- Briefly, you need to add the correct Security Groups: https://docs.csc.fi/cloud/pouta/security/#restrictive-firewall-white-listing - but for that you need to figure out what are the IP ranges for UoH - perhaps ask the UoH IT support

- **Q211** Are VTT people allowed to use this academic Matlab licence at CSC? We do have our own licenses too, but I doubt they won't be that easy to forward to CSC environment. We VTTers are kind of academic too... / Harri
- I believe that VTT is not allowed to use CSC's MATLAB licenses, unfortunately. Does VTT have license server and does it allow traffic from Puhti?
- >> I'm quite sure we have some sort of centralized Matlab license system at VTT, but I have no idea how it works. This kind of approach should probably be handled via our IT support anyways.
- In Bash setting environment variable like `export MLM_LICENSE_FILE=[e-mail]` should be set to point to the license server. We need to figure out the port number and domain name, and whether the server allows traffic from Puhti domains.
- IF VTT servers allows *any* unauthenticated matlab from Puhti THEN keep the port@domain secret ... / Jukka Lehtonen

- **Q212** In my cPouta instance, I have a flavor of Ubuntu 22 with 4 GPUs each of which containing Tesla P100-PCIE-16GB. 16GB is quite insufficent for my application. I was wondering if there might be another flavor which could offer 4 GPUs of 32GB, probably Tesla V100 or any other equivalent?
- In https://docs.csc.fi/cloud/pouta/vm-flavors-and-billing/#gpu-flavors you can see all the available flavors.
- If HPC/supercomputer is an option for you (instead of Pouta), our Puhti, Mahti and LUMI supercomputers also have GPUs with more memory.

- **Q213** Is it posible to extend storage in puhti? In our team the scratch storge almost full (I can not longer to do over it). Can the project team request increase to the project’s disk quota in MyCSC service? or this can be done by project manager? + Yohannes /
- You can request an increase to the project's disk quota in MyCSC. See instructions here: https://docs.csc.fi/accounts/how-to-increase-disk-quotas/

- **Q214** Is it possible to install deep learning tool box? If yes, please explain the procedure/ Usman Muhammad

- **Q215** How to enable the Add-On explorer so one can install the tool box by own requirement / Usman Muhammad

- **Q216** when installing a package using either pip insall package_name or conda install package_name using tykky module, I do the following:
- module purge
- module load tykky
- cd /projappl/project_XXXXX
- modify packages.sh file as needed! pip install or conda install ...
- conda-containerize update CondaCSC --post-install packages.sh
- export PATH="/projappl/project_XXXX/CondaCSC/bin:$PATH"
I could install my packages using the aforementioned instruction, problem on the other hand is that this is significantly time-consuming even for 1 package.

[ INFO ] Creating sqfs image 
Parallel mksquashfs: Using 8 processors
Creating 4.0 filesystem on _deploy/img.sqfs, block size 131072.
[========================== takes forever to finish.. :(

Am I doing something wrong? or is there any additional step to consider to speedup such installation?

- Tykky is creating a whole new container which is quite a big job. For small package installations you might be able to install to a virtual environment for example with `pip install` - it depends on the case.

But I thought one should completely avoid using ```conda create -n py39 python=3.9 -y``` for instance (creating a virtual env) and we were highly recommended to use tykki container instead according to the puhti doc page!
- The answer: you can create the base environment with tykky but simple package additions can be done with `pip install --user` or using Python venv: https://docs.csc.fi/apps/python/#installing-python-packages-to-existing-modules

- **Q217** How could I use 4 GPUs parellel? + Mohamed / nick name

- Do you mean for Matlab or some other case?
- No, in Puhti it is possible to use 4 GPUs in parallel. How I could do that? and How could I use the other disk instead of shared filesystem ?
- This was discussed in the breakout room, but some basic documentation links here:
- https://docs.csc.fi/computing/running/getting-started/
- https://docs.csc.fi/support/tutorials/ml-multi/
- https://docs.csc.fi/support/tutorials/ml-data/#fast-local-drive-puhti-and-mahti-only

## 2024-02-21 session

**Short talk today:** Troubleshooting in CSC computing environment ([Slides](https://a3s.fi/saren-2001659-pub/Troubleshooting_in_CSC_HPC_environment_2024-02-21.pdf))

**Short talk next week (28.2.):** Using Matlab on CSC supercomputer Puhti

**Advertisements:**

- Free hybrid seminar on [Machine Learning with spatial data](https://ssl.eventilla.com/geoml_seminar) on 27.2.
- [Introduction to Geospatial Python](https://ssl.eventilla.com/geospatial_python) on 4.-6.3, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6.-4.7., registration open.

### Zoom breakout rooms and topics

Room 1 - Questions on short talk + Getting started with CSC services (Ari-Matti)
Room 2 - Q207 Snakemake (Laxmana)
Room 3 -
Room 4 - Q209 MLFlow (Rahti people)
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q207** I would like some help with running my Snakemake pipeline (how to call modules in my Snakefile? How to run jobs in parallel?) - Natalie van Dis

- Load puhti modules in batch script before executing snakemake command or build tools in Tykky container or singularity container.
- At least one way to do is to request more resources in batch directive and setting parameters in Hyperqueue executor can parallelise your workflow.

- **Q208** I would like some help with the generation of container using apptainer / Sébastien Le Maguer
I am mainly stuck with some size limit which are a bit cryptic for me

- Try setting up $TMPDIR/CACHEDIR explicilty to avoid disc space errors before start building container:

    export APPTAINER_TMPDIR=$LOCAL_SCRATCH    # Set the local storage area to the environment.. 
    export APPTAINER_CACHEDIR=$LOCAL_SCRATCH  

- **Q209** (CSC #683173) Mlflow service started from the template in the Rahti dashboard hangs at 'Pending'. - Nazaal.
- room 4 with Rahti team (deployment issue)
- Issue was resolved with a new Rahti project and deploying the environment there.

## 2024-02-14 session

**Short talk:** Chipster - easy-to-use bioinformatics analysis software for high-throughput data
- Chipster course materials: https://chipster.rahtiapp.fi/manual/courses.html
- Chipster website: https://chipster.csc.fi
-Chipster user interface (go ahead and try with your haka/virtu account): https://chipster.rahtiapp.fi/home

**Advertisements:**

- Free hybrid seminar on [Machine Learning with spatial data](https://ssl.eventilla.com/geoml_seminar) on 27.2
- [Introduction to Geospatial Python](https://ssl.eventilla.com/geospatial_python) on 4.-6.3, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6-4.7, registration open.

### Zoom breakout rooms and topics

Room 1 - Questions on presentation (Chipster & Eija)
Room 2 - Getting started with CSC services (Laxmana)
Room 3 - Kimmo / SD Desktop
Room 4 - ssh problem /Ari-Matti
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q205** Are there any tutorials on how to access and use Alphafold? + sanjana
- https://docs.csc.fi/apps/alphafold/
- Regrettably we don't have anything more detailed on Alphafold itself. You will need to find some instructions or tutorials on the net.

- **Q206** logging into Puhti with ssh (breakout room 4)

## 2024-02-07 session

**Short talk:** CSC cloud services for education

**Advertisements:**

- [High-level GPU Programming](https://ssl.eventilla.com/hlgpup) 14-16.2, registration closing soon! LAST CHANCE to REGISTER!
- Free hybrid seminar on [Machine Learning with spatial data](https://ssl.eventilla.com/geoml_seminar) on 27.2
- [Introduction to Geospatial Python](https://ssl.eventilla.com/geospatial_python) on 4.-6.3, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6-4.7, registration open.

### Zoom breakout rooms and topics

Room 1 - Questions on presentation (Joona)
Room 2 - Getting started with CSC services (samantha)
Room 3 - Support session for "CSC Computing Environment (basics level)" course (Rasmus)
Room 4 - Q203
Room 5 -
Room 6 -
Room 7 -
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q202** Lukashevskyi Arsenii (arseluka) Hello, we have a morphological tagger for texts in Ukrainian. It is written in Groovy and so we can't run this program on Puhti because we don't have the necessary libraries. It is a language similar to Java, but unfortunately rewriting this program will take considerable time. How do we install Groovy on our project? Thank you in advance for your help.
**A**: We haven't installed Groovy on Puhti, but it seems to be easy to do it yourself:
1. download groovy zip from here: https://groovy.apache.org/download.html
2. unzip on puhti
3. module load biojava
4. now groovy commands can be found inside the extracted folder in `bin` directory

- **Q203** I have an minio instance running on Rahti, I want to use it as an artifact store in Mlflow so I can access the artifacts easily from any Mlflow client, since right now the artifact URI depends on the machine which the experiment is run on. I am using Postgres running on Rahti for the tracking URI. I can do some hacky things to solve get artifact URIs from other clients but would prefer to use the minio instance. Some links that seem relevant while I was trying to solve this problem:
- https://mlflow.org/docs/latest/tracking/artifacts-stores.html
- https://blog.min.io/setting-up-a-development-machine-with-mlflow-and-minio/
- https://github.com/mlflow/mlflow/issues/5514
Nazaal

- **Q204** I have an error with Slurm where the same script doesn't run on Puhti (gives slurm_load_jobs error: Invalid job id specified) but runs on Mahti. Nazaal.
- A: Hard to say without seeing your batch script. I assume you are trying to query some data for a job that has been run on Mahti? That will not work on Puhti as the Slurm accounting databases are separate.

## 2024-01-31 session

**Short talk:** How to write a support request [(Slides PDF)](https://a3s.fi/saren-2001659-pub/How_to_write_a_support_request_2024-01-31.pdf
)

**Advertisements:**
- [CSC Computing Environment course](https://ssl.eventilla.com/event/M0azg), 7.-8.2. Course is fully booked, but you can register to the waiting list!
- [High-level GPU Programming](https://ssl.eventilla.com/hlgpup) 14-16.2, registration closing soon! LAST CHANCE to REGISTER!
- Free hybrid seminar on [Machine Learning with spatial data](https://ssl.eventilla.com/geoml_seminar) on 27.2
- [Introduction to Geospatial Python](https://ssl.eventilla.com/geospatial_python) on 4.-6.3, registration open.
- [HPC Summer School at CSC](https://ssl.eventilla.com/event/summerschool2024) 25.6-4.7, registration open.

### Zoom breakout rooms and topics

Room 1 - Getting started with CSC services (Ari-Matti)
Room 2 - Pre-course support session for "CSC Computing Environment (basics level)" course
Room 3 - Chipster question (**private room**) (Ellimari)
Room 4 -
Room 5 - Rahti and/or Pouta Room (Q200)
Room 6 - Q197 + Q201 (Mats)()
Room 7 - Matlab (Jaan)
Room 8 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q197** Hi, I need advice on how to organize an image analysis pipeline/storage optimally (academic use, UH). I am superficially familiar with CSC services after the "Using CSC computing environment efficiently" course, but not confident enough to start right away. I was thinking about using Allas for 'accessible' storage, and Jupyter (on Puhti?) for downstream analysis. I would greatly appreciate any tips and tricks. - Taras
- room 6
- A: https://docs.csc.fi/data/Allas/introduction/ Allas file system, S3 vs Swift
- A: https://my.csc.fi how to apply for a project

- **Q198** Chipster question related to users data - Anna K
- room 3

- **Q199** Hi, I need to move all the data on Allas from a project that is expiring to another project. Is there a simple way to do this? I know how to move the files from Allas to scratch, and then from scratch to the other project. But I'm wondering if there is a way to directly move all the project contents to another project. - Tia
- A:

- **Q200** Rahti related issue. I need to discuss about running docker file of shiny app at rahti. I have some error to run it "error: open /tmp/build/inputs/Dockerfile: no such file or directory". + Umair Seemab / Umair
- room 5
- A: First thing, the Dockerfile must be with a capital D to solve the error "error: open /tmp/build/inputs/Dockerfile: no such file or directory". Then, the other error "Error response from daemon: missing signature key" is because the format of the image from DockerHub is not oci-compatible docker. Skopeo can be used (https://docs.csc.fi/support/faq/get-image-format/#workarounds) to copy the image from the DockerHub to the Rahti registry of the project. It will convert the image.

- **Q201** Hi, I have a postgres instance running on Mahti. I am running this to save experimental results from Mlflow (Python package). However, I am unable to connect to it from there, the connection times out. I suspect the IP address is wrong? I got that IP via (after logging in via oc) "oc get pods -o wide" then "oc exec -it PODNAME -- bash" then "hostname -i". + Nazaal Ibrahim / Nazaal
- room 6
- A:

## 2024-01-24 session

**Short talk:** Which CSC service for research suits my needs? -> [Materials](https://siili.rahtiapp.fi/weekly_CSC_service_overview#)

### Zoom breakout rooms and topics

Room 1 - Getting started with CSC services
Room 2 - Questions related to Rahti (Joona)
Room 3 - Mats
Room 4 - Connecting to Puhti (Samantha)

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q192** When using Funet for hosting my dataset, how can I assign a permanent link and others using the `wget` command to download it in their own directory in terminal? (is that even possible?) + your name / nick name

A: if you refer to Funet Filesender then it's not really meant for permanent hosting of data, but rather sending a file to someone else. For project life-time hosting Allas is better suited, or for longer term hosting see the [list on Samantha's page](https://siili.rahtiapp.fi/weekly_CSC_service_overview#Data-services)

- My dataset is sized more than 250GB and would like to use it in my Rahti web interface when saved as PVC, so ALLAS is not working for Rahti!

- **Q193** Is there any service to host the shiny app like shinyapps.io just to test the developed shinyapp? / Umair

A: short answer is no, we have no specific service like that. Of course you can probably run your shiny app in Pouta or Rahti, but that requires a bit more work from your side. If you have more details, we can maybe help you get started.
- having a dockerfile is a good starting point for making use of Rahti
- docker uses root user level, Rahti is rootless container -> might generate some hiccups, but can be solved, you can find possible solutions, eg making it run rootless in Rahti documentation: https://docs.csc.fi/cloud/rahti/rahti-what-is/

- **Q194** Matlab question / Siiri R.
- A: Sorry Siiri, our Matlab specialists are busy elsewhere at the moment :/ Best for you to write a support request = email your question to [e-mail] (the best specialist available will then reply to you!)

- **Q195** I got the question concerining Allas. I know that all the files have an individual URL-address. But if I send that link to collaborating person that does not have any connection to CSC, can that person download the file?/ APK
- If the file is marked as public then it can be accessed from anywhere.
- You can make whole buckets public eg from the Pouta web interface (or with CLI tools like s3cmd), so that everyone with that link can access your files in that bucket without the need for a CSC account

- **Q196** Ssh problem / Mira
- A: It was this one, and following the instructions here solved the issue: https://docs.csc.fi/support/faq/i-cannot-login/#why-is-my-ssh-client-saying-corrupted-mac-on-input

## 2024-01-17 session

**Short talk:** Machine learning on supercomputers, Part 2: Workflows for using Jupyter Notebooks and GPUs on Puhti [(slides PDF)](https://a3s.fi/mats/gpus-and-notebooks.pdf)

### Zoom breakout rooms and topics

Room 1 - Getting started with CSC services (Ari-Matti)
Room 2 - Machine learning / Short talk discussions (Mats)
Room 3 - Rahti (Jemal)
Room 4 - Allas (Kimmo)
Room 5 - Matlab (Jaan)
Room 6 - Student project (Laxman)
Room 7 - GPU program (Cristian)

### Questions

- **Q188** Hi, could you please point to CSC docs section or tutorial explaining how to install toolboxes (I need Image Processing Toolbox) to use in Matlab web interface on Puhti? I don't know Matlab, just need to test run few scripts before rewriting them in python. Thank you!
- Unfortunately, we don't have a license for the Image Processing Toolbox, thus we don't have it installed. Your best option is to install matlab with the toolbox on your own computer using a campus license (assuming you are a student).
- thanks, got it
- I am here without the mic.. But I've got my answer, thank you, no need to make a room for that
- Great!

- **Q189** How can I create a student project? I'm taking a course by our university on programming supercomputers, and that course instructs to create a student project. However, as a part of a research group, I already have access to a 'proper' project in MyCSC and when I try to create a new project, I only have options to create an Academic/Commercial/Course/LUMI project, not a student one. There's no option to create a student project like in the video in docs, and the Course project type does not seem to be the correct one. -ET
- Please see: https://docs.csc.fi/accounts/how-to-create-new-project/#student and https://docs.csc.fi/support/tutorials/student_quick/
- those are the pages that I checked. When I press "New project" in the projects page, I don't get the form for creating a student project, but a form with fields for the project name and description, and radio buttons for project type (Academic/Commercial/Course/LUMI). Do I need to create a new CSC account?
- I asked for further instructions from our customer account people, but have not received an answer yet. Please send mail to servicedesk@csc,fi so we can put you in contact with the correct people.

- **Q190** I have a very simple GPU program that tests more or less random access performance and the kernel takes around 23 times longer on LUMI than on Mahti, I would like to figure out if maybe I'm missing something with the compilation of the program on LUMI/Mahti, or if it's an underlying hardware limitation - Anton Bogun

- **Q191** Is papermill available in Puhti, Mahti and LUMI?
- Yes, it should be preinstalled at least in CSC's PyTorch and TensorFlow modules in all three systems. It is currently not available in the python-data module, but it can be easily installed for example with `pip install --user papermill`. (I will add it to future versions of python-data.)

## 2024-01-10 session

**Short talk:** Machine learning on supercomputers, Part 1: getting started and using GPUs

**Advertisements:**
- [CSC Computing Environment course](https://ssl.eventilla.com/event/M0azg), 7.-8.2., registration open!
- [Practical Deep Learning course](https://ssl.eventilla.com/event/bJbGW), 8.-9.2, registration open!
- [High-level GPU Programming](https://ssl.eventilla.com/hlgpup) 14-16.2, registration open!
- Free hybrid seminar on [Machine Learning with spatial data](https://ssl.eventilla.com/geoml_seminar) on 27.02.24
- HPC Summer School at CSC 25.06-04.07, registration opens soon.

### Zoom breakout rooms and topics

Room 1 - Getting started with CSC services (Samantha)
Room 2 - Machine learning (Mats)
Room 3 - Rahti questions (Jemal)
Room 4 - SD Desktop (M)
Room 5 - Matlab (Jaan)
Room 6 -
Room 7 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
Previous questions and answers can be found in the archives, see links section above.
:::

- **Q178** I have a remote disk mount set-up on my Mac. I am unable to actually view the contents inside my scratch and projappl folders. How does this work? - Kavya
A: https://docs.csc.fi/data/moving/disk_mount/
Not solved by the info in the link, advised to send an email to [e-mail].

- **Q179** is CSC going to update CUDA 12.0 to a newer version quite soon?

$ nvidia-smi | grep "CUDA Version"
| NVIDIA-SMI 525.125.06   Driver Version: 525.125.06   CUDA Version: 12.0     |

I use Jaxlib and get this warning/error for slow execusion due to NVIDIA driver: **The NVIDIA driver's CUDA version is 12.0 which is older than the ptxas CUDA version (12.3.107). Because the driver is older than the ptxas version, XLA is disabling parallel compilation, which may slow down compilation. You should update your NVIDIA driver or use the NVIDIA-provided CUDA forward compatibility packages.** + Farid
- A: We can ask our sysadmins if the CUDA drivers can be updated, but it might not be possible. Can you email [e-mail] and include a small testable example so that we can work on finding a work-around?

- **Q180** Given:

# file named as puhti_int_gpu.sh
#!/bin/bash

srun -J v100_gputest --account=project_XXXXX --partition=gputest --gres=gpu:v100:1 --time=0-00:15:00 --mem=256G --ntasks=1 --cpus-per-task=10 --pty /bin/bash -i
module load git

then in terminal:

$ source puhti_int_gpu.sh
$ git --version
bash: git: command not found!

if I ```$ exit ``` and redo

$ source puhti_int_gpu.sh
$ git --version
git version 2.35.2

I am confused! why? +Farid

- A: I'm not entirely sure what you are trying to do? Best way to start an interactive session is to use the `sinteractive` command.

- **Q181** (from chat) When running a GPU job, is the GPU dedicated to that job, such that the GPU metrics (energy, memory etc) observed are specific to that job?
- resolved in talk

- **Q182** (from chat) Hi, I wish to know, number of workers mean no of CPUs or cores in a GPU.
- resolved in talk

- **Q183-187** (from getting started breakoutroom as reminder) SD Desktop like solution for Uni Oulu? What is CSC? How to activate conda from within tykky environment? How to choose resources to request when starting RStudio?

## 2024-01-03 session

**Short talk:** New features in docs.csc.fi

**Advertisement:** Next [CSC Computing Environment course](https://ssl.eventilla.com/event/M0azg) 7.-8.2. registration open now

### Questions

- **Q173** How to use Snakemake with HyperQueue (Tommi Vatanen/Freed Ahmad/Shuai Ji)? We would like to understand important parameters to implement massively parallel analysis workflows.
- One can use similar hyperqueue code as part of snakemake. We have related examples (nextflow or other cases) Please check our tutorials: https://csc-training.github.io/csc-env-eff/hands-on/throughput/hyperqueue.html and https://docs.csc.fi/support/tutorials/nextflow-hq/

- **Q174** Rahti | Pod practical meaning? Let's say I allocate 9 pods to my openshift container platform and each of which comes with 16GiB memory (x4 cpus). Does this essentially mean I allocated 9 * 16GiB = 144GiB to my container (project)? + Farid
- Here are some links about what we've discussed:
- https://docs.csc.fi/cloud/rahti/usage/projects_and_quota/
- https://docs.csc.fi/cloud/rahti/concepts/#kubernetes-concepts
- https://kubernetes.io/docs/concepts/workloads/controllers/deployment/
- https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/

- **Q175** Hello, i want to start using VASP software and i Don't know how to access via CSC computers?! +(Nick)name
- https://docs.csc.fi/apps/vasp/ -> `module load vasp` on Puhti
- register your license, send a message to servicedesk
- examples on docs page

- **Q176** When we want to apply for a new resource, it asks us to add some publications. What we should do if the project is new and there are no publications yet? +(Nick)name
- Some older or related publications are also ok, could be from anyone in the project
- If you do not have any yet, you need to send a message to [e-mail]

- **Q177** Using LUMI AMD GPU, strange SQLite errors +Kalle
- MIopen cache-related environment variables need to be set, see example: https://docs.lumi-supercomputer.eu/software/packages/pytorch/#example
- Note also that we have a module for PyTorch which has that automatically set (if that works for you): https://docs.csc.fi/apps/pytorch/

## 2023-12-20 session

**Short talk:** Research Data Management online course, RDMkit and FAIR principles videos
* 18.1.2024 at 10:00-11.30 How to use the Allas Storage Service? https://ssl.eventilla.com/event/WLOaO
* 8.2.2024 at 10:00-12:00 Research data management and data documentation in long-term research projects https://ssl.eventilla.com/event/wJed3
* elearning course: https://e-learn.csc.fi/enrol/index.php?id=63
* playlist: https://video.csc.fi/playlist/dedicated/0_723o7br1/0_q2jf8r0y
* RDMKit: https://rdmkit.elixir-europe.org/

### Questions

- **Q171** I want to use custom made apps of Matlab in Puhti (using the virtual desktop). I was told this is not possible and it is better I use cPouta. The message I got was: " to create your own MATLAB installation on cPouta cloud service and run your interactive MATLAB sessions there"... How does that really work? I am a new user and I am just getting familiar with Puhti. My specific question is: how do I "create my own MATLAB installation"? +Carlos Aguilar
- Basic Pouta documentation: https://docs.csc.fi/cloud/pouta/
- This is quite detailed instructions how to get a linux up and running in Pouta https://docs.csc.fi/cloud/pouta/launch-vm-from-web-gui/
- After getting the machine up and running - it's basically just a linux over SSH and there are plethora of information about it.
- elearning course material: https://e-learn.csc.fi/course/view.php?id=102

- **Q172** Getting started with LUMI +Pankaj
- https://lumi-supercomputer.eu/get-started/
- https://docs.lumi-supercomputer.eu/helpdesk/
- LUMI support sessions: https://www.lumi-supercomputer.eu/events/usercoffeebreaks/

## 2023-12-13 session

**Short talk:** Installing own R packages on Puhti

### Zoom breakout rooms and topics

Room 1 - R installations (Heli)
Room 2 - Q169: MySQLdb -> Kimmo in main room
Room 3 - Q170: model fitting on GPU (Mats)
Room 4 -
Room 5 -

### Questions

- **Q169** I am using an app within a Conda environment (on Puhti) and it uses MySQLdb to export results and I am currently unable to do this. Based on my research, there seems to be an incompatibility between the MySQLdb version used by the app and the one in Puhti's environment. How do I circumvent this? - Kavya
->

- **Q170** I have a large model fitting task, which according to my evaluation with short test runs, would last over 72 hours in a GPU node. Do I get it right that the only way to execute such model fitting is to store some intermediate output alongside my computation, and split my task into a cascade of sequentially placed jobs, where each next one is initialized by the result of the prevuious one? I am interested in my options both for Puhti and LUMI. - Gleb

## 2023-11-29 session

**Short talk:** Using R on supercomputer Puhti

### Zoom breakout rooms and topics

Room 1 - R on supercomputer - Heli
Room 2 - CSC Computing environment course - Maria and friends (Kimmo, Ari-Matti, Laxman, Rasmus, Nino)
Room 3 - Q165 : best practices for git on supercomputer - (Samantha if no one else wants to)
Room 4 - Q166: Snakemake - Laxamana
Room 5 - Q167: GPU - Mats?
Room 7 -

### Questions

- **Q165** Using git on Puhti. It merges our sessions since the project is cloned in /projappl +Khizar
- So you would like to have a project cloned twice/multiple times by different users (to make use of different branches) in projappl?
- Each team member can also clone it in their home directory, but I was hoping for a cleaner solution.
- One possibility would be to give the local clone different names `git clone <address> /projappl/path/to/own/clone` so that every team would work in their own directory? Would that help already?
- Yes, but then the solution is ultimately a clone for each team member, be it in their home directories or within separate sub-directories in /projappl. Thank you for the suggestion though, it does seems like it will solve the issue.
- We can still discuss in a breakoutroom after the short talk :)
- I think this is how git is meant to be used: each user has their own clone, and collaboration happens when pushing the commits to a central location (like github).

- **Q166** Snakemake package management when conda is not advisable in puhti +Pande
- Specific question: the snakemake environment is already available in puhti, but some packages needed in the pipeline are half available (can be called using --use-envmodules), but some are not available that needed to be configure for example through conda by specifying env.yaml file. What is the best practice for this that you can suggest? Note: apptainer is my last resources preference.

- **Q167** It seems to me that increasing the number of GPU nodes, in Puhti, does not necessary increase the speed of the computation. Also what about the time limitation, i think 72 houres is the maximum. What if mine needs far more than that time?! + Ali Saghi
- What kind of computation are you running? Is this related to machine learning? Yes
- Your program needs to explicitly support multi-GPU and/or multi-node jobs. It is not automatic. What program or framework are you using? Tensorflow/Keras
- tf.distribute.MirroredStrategy or Horovod are the most common frameworks for TF/Keras. MirroredStrategy is the easiest to start with, but it only supports multiple GPUs in one node (so max 4 GPUs in Puhti).
- For training jobs lasting more than 72 hours a common approach is to use snapshots (so having a snapshot-saving callback in Keras, i.e. https://keras.io/api/callbacks/model_checkpoint/) and then chain multiple jobs that in the beginning load the most recent snapshot of the model and continue from there.
- Ali: You can also join breakout room 5 for more discussion (if needed)
- There is a Practical Deep Learning course coming up at CSC, probably in February, keep an eye on https://csc.fi/training

- **Q168** How to Access VASP via CSC +sameur23
- You need to bring your own license (CSC does not have one). If you have your own license, send an email to <[e-mail]> and we will add you to the VASP user group which gives you access to the CSC VASP installation.

## 2023-11-22 session

**Short talk:** Updates in the Pouta cloud service
* Rocky update: https://docs.csc.fi/cloud/pouta/application-dev/
(You can find a lot of information on Pouta also in our documentation: https://docs.csc.fi/cloud/pouta/)

### Zoom breakout rooms and topics

Room 1 - Pouta
Room 2 - "[CSC Computing Environment](https://ssl.eventilla.com/event/1N0g8)" pre-course support session
Room 3 - Question 164 - ML and containers

### Questions

- **Q164** when setting up a container, I have to use .sif file. In the webite, there is an example using sinteractive. This time should be the same as the one in sh file? Also there is an option to use --nv in oder to use GPU. If we don't include, what is the result? I want to train a LSTM model in multi GPUs but a colleague advised me to use containers in order to set non-zero reccurent_dropout and run with the fast cuDNN kernel. +Panagiotis
- room 3 with Mats

## 2023-10-25 session

Short talk on Python installations on supercomputer

- **Q161** Curiosity: Schrödinger Maestro, local install (on AlmaLinux 9). Show molecule with semi-transparent surface. "Save Image", PNG with transparency.
- On Maestro 2023.2 all is good
- On Maestro 2023.3 there are rectangular regions on the surface, non-smooth mess
- Both installs were made the same way, so the bits in the package are to be suspected
- We will report to Schrödinger, but anybody else seen similar? *Jukka Lehtonen*
- **Answer:** Have not seen this myself or heard others reporting this. I have noticed, however, that 2023.X versions have started to warn from time to time about missing graphics libraries on Puhti (which is not an issue since the GUI is best not used on the supercomputer anyway). But clearly, Schrödinger has likely made some changes here that might affect how Maestro behaves graphics-wise. Contacting them directly is indeed the best way forward.

- **Q162** What's the best way to set up a Python environment for a workshop/course? (many people needing the same somewhat custom environment.) I'd like to avoid having everyone needing to run a set of pip install at the same time when the course starts. -Max V
- Also wondering if this is a good use case for Tykky, since the environment needed is in an Apptainer image but relies on some pip installs in the system
- This is a course that needs to run on the supercomputer? Is it using Jupyter?
- A: Yes, it's a tutorial running on the cluster. It's a (set of) scripts at the moment but can be adapted to Jupyter.
- No need to do it in Jupyter if it is not needed by default :)
- If you have a course project you can have a common installation in the project's projappl, for example a virtualenv.
- Ok, that could be a good solution. I was a little worried about exceeding the number-of-files limit, but if it's one installation for the course (rather than one per user) this is a lot less severe.
- Yes, exactly. Of course the common installation can also be with tykky, and then it's even better for the file system.

- **Q163** I used python and I was able to access Allas (https://docs.csc.fi/data/Allas/using_allas/python_library/) but I need to fetch the data from Allas to display it on table on a website.One method I used is that I created Flask app and tried to access it. It seems not to work. What do you advice me?! Jeba
- Flask currently run locally, later plan to move to e.g. Pouta
- generally possible e.g. via s3
- -> should work, but ends in 404
- -> (explanation in session) sounds like an issue with the Python code

## 2023-09-20 session

**Short talk:** no short talk today, DARIAH-FI/FIN-CLARIAH people will be joining to ask questions - although any CSC user free to ask questions as usual

### Zoom breakout rooms and topics

Room 1 - DARIAH-FI related questions
Room 2 - srun step creation still disabled-issue, Mats
Room 3 - Allas file transfers, Mats
Room 4 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q152** What is the upper threshold for the Ram memory and CPU cores that we can ask for in Rahti? + Mehrdad from DARIAH-FI
- A: This is decided case by case. User needs to create a ticket to service desk adding an explanation of the need. Rahti's PO decides. The upper limit for a single pod is around 40GiB of memory and 15 cores. he higher you go, the more difficult your pod scheduling will be and you may end up waiting for a free slot.

- **Q153** Is there any document on how to use the "metrics" section of a pod on Rahti? + Mehrdad from DARIAH-FI
- A: This is broken in current´s version of Rahti.
- (new version of Rahti solves this, probably coming end of this year)

- **Q154** Let's say I would like to run Python code within my R scripts with reticulate package. The environment would be Rstudio in Rahti. I have run into trouble when I try to install Python dependencies from the terminal. Is this not possible due to insufficient user permissions etc? If this is not possible, could I run R code and Python code in a single Jupyter notebook by creating a Tykky container wrapper? Pyry from DARIAH-FI
- A: Rahti does not allow installing software in running images. One way to solve this is to make your own image using the current one as a base.
- A: Tykky is for Puhti/Mahti/LUMI only. Would it make sense to move the use from Rahti to supercomputers?

- **Q155** We would like to offer large dataset (about 10 Tb) from within allas? How to set this up, what does it require? Imagine that we do not have project in CSC so starting from scratch. -- TuulaP from Dariah-FI
- A: [Set up a project](https://docs.csc.fi/accounts/how-to-create-new-project/). [Add Allas service](https://docs.csc.fi/data/Allas/accessing_allas/#gaining-access) for your project. The default quota happens to be 10 Tb, but requests for more quota are generally accepted within reason. If your data is not already on a CSC machine, but somewhere remote, it's best to upload it directly to Allas with swift or [s3](https://docs.csc.fi/data/Allas/using_allas/s3_client/) tools. (On the other hand, if it's already on a CSC HPC machine, you can use a-tools, like [a-publish](https://docs.csc.fi/data/Allas/using_allas/a_commands/#a-publish) to upload the data to a public bucket in Allas.) A file uploaded to a public bucket can be accessed with URL: https://a3s.fi/bucket-name/object-name
- S3 client (or swift, or whatever client you prefer) can be installed on your system and the data can be uploaded from there. [docs.csc.fi](https://docs.csc.fi/data/Allas/using_allas/s3_client/) has some pointers. They link to a script called `allas_conf` which can handle setting up the connection for you.
- What if the researcher does not use Puhti and has Windows in use locally?
- Files can be uploaded via the [Allas web UI](https://allas.csc.fi) too (sharing buckets is also possible) (not feasible for terabyte scale though)
- [cyberduck](https://docs.csc.fi/data/Allas/using_allas/cyberduck/) is an option

- **Q156** From researcher angle, if I am a researcher wanting to use a large dataset from allas, what should I do? (see Q155) -- TuulaP from Dariah-Fi
- A: just use the URL mentioned in Q155 to download the data
- What if there is tens or hundreds of files?

- **Q157** If a large dataset exists in allas, then how long can it be expected to remain there? (see Q155) -- TuulaP from Dariah-Fi
- A: data can be kept in Allas as long as the CSC project is active. CSC projects are expected to last 3-5 years in general

- **Q158** Follow-up question to Q154: How should I proceed with creating my own image using some other image as a base? Pyry from DARIAH-FI
- A: We have a guide at: https://docs.csc.fi/cloud/rahti/images/creating/

- **Q159** Follow-up question to Q158: It is stated in CSC docs that "Rahti has a very limited amount of A100/P100 GPUs" and "Rahti projects will not have access to GPU by default". The need for using Python in the first place is related to being able to use GPU acceleration for certain tasks. Would some other environment than Rahti, eg Pouta offer the option to run Python as well as R simultaneously?
- Pouta has a some GPUs available (also very limited, but a bit easier to access than Rahti). For big GPU resources, use the supercomputers: Puhti, Mahti or LUMI. Puhti has R, RStudio and Python available.

## 2023-09-13 session

:::info
**Quick info: CSC capacity survey** - the renewal of the national data management and computing environment (DL2026).

This survey is meant for CSC's current and potential users and stakeholders with interest in developing the national supercomputing, cloud and related data storage.

Please, fill in the survey by 17.9.2023: https://link.webropolsurveys.com/S/D8991442531EE29E
:::

**Short talk: Benefits of version control in HPC environment**

Some links and material about GitHub
- [Comic reasoning why you should use version control](https://phdcomics.com/comics/archive.php?comicid=1531)
- [CodeRefinery-lesson: GitHub without command line](https://coderefinery.github.io/github-without-command-line/basics/#)
- [CodeRefinery-lesson: Git intro](https://coderefinery.github.io/git-intro/)
- [CodeRefinery-lesson: Jupyter (inlcudes version control)](https://coderefinery.github.io/jupyter/version-control/)

**Upcoming courses:**
- [Free CodeRefinery workshop: version control and reproducible research](https://coderefinery.github.io/2023-09-19-workshop/) September 19-21 and 26-28
- [All CSC trainings](https://www.csc.fi/training)

### Zoom breakout rooms and topics

Room 1 - CSC Computing environment course support session
Room 2 - Puhti installations
Room 3 -
Room 4 - Git questions (Matias)

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q159** I'd like to have specific help with program installation. +Ina
- **Q160** Also specific help with program install on puhti. +Patrick

- Two Python-installation questions, some issues related to environments being containerized.

## 2023-09-06 session

Short talk: What are the private and public keys in SD Connect? - Asymmetric encryption with crypt4gh
+ quick reminder about CSC capacity survey

### Questions

- Discussion in the main room regarding encryption short talk topics

## 2023-08-30 session

Short talk: **No short talk today**

Some links and material about GitHub
- [Comic reasoning why you should use version control](https://phdcomics.com/comics/archive.php?comicid=1531)
- [CodeRefinery: GitHub without command line](https://coderefinery.github.io/github-without-command-line/basics/#)
- [CodeRefinery: Git intro](https://coderefinery.github.io/git-intro/)

Upcoming courses:
- [Using CSC Computing environment](https://ssl.eventilla.com/enveff2023sep) 19.-21.9. -fully booked, register to the waitlist!
- [Code refinery: version control and reproducible research](https://coderefinery.github.io/2023-09-19-workshop/) September 19-21 and 26-28
- [All trainings](https://www.csc.fi/training)

### Questions

- No questions this time

## 2023-08-23 session

Short talk: **MyCSC: Course project type**

Upcoming courses:
- [Using CSC Computing environment](https://ssl.eventilla.com/enveff2023sep)
- [All trainings](https://www.csc.fi/training)

### Zoom breakout rooms and topics

Room 1 - Puhti web interface & RStudio (Mirte & Heli)
Room 2 - SD Desktop (Felipe & Kimmo)
Room 3 -
Room 4 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q149** How to use Puhti webinterface (Rstudio)? I try to run an Rscript in this webinterface but run into some problems. Mirte
- Discussed what kind of jobs RStudio in the web interface is suitable (light to medium jobs, not very heavy ones), R package installation instructions: https://docs.csc.fi/apps/r-env/#r-package-installations

- **Q150** Is it possible to process data in puhti/mahti that comes from SD Connect? Otherwise, is SD desktop providing gpu computing? - Felipe

- **Q151** Will students (university of Helsinki) need course accounts or will their HU accounts work at CSC? Liisa
- Yes, they can use their HU accounts via Haka login
- Does it work with ssh?
**A:** Through web interface yes (web interface supports Haka login) but if you have a separate service with ssh login (e.g. a virtual machine in cPouta) that service would need its own user authentication method. cPouta supports the ssh-keypair authentication method.

## 2023-08-16 session

Short talk: [The batch job system in CSC's HPC environment](https://a3s.fi/CSC_training/05_batch_jobs.html)
See also slides from [HPC crash course, part 1](https://a3s.fi/mlehtiva-2001659-pub/mini-crash-course-batch-job.pdf)

Upcoming courses:
[Using CSC Computing environment](https://ssl.eventilla.com/enveff2023sep)
[All trainings](https://www.csc.fi/training)

### Zoom breakout rooms and topics

Room 1 - Jupyter Notebooks in Puhti & Rahti for course (Haining & Otto, Ian & Laxmana, Mats)
Room 2 - Creating allas buckets (Kim & Kimmo) + bonus, encryption keys!

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q144** When running Jupyter notebooks on Puhti, sometimes the codeblocks won't run. An asterisk appears on the left of the codeblock indicating that it's running but nothing happens. Also, the bottom left of the screen says 'Idle' instead of 'Busy'. What could be causing this issue? Restarting the kernel sometimes helps but not reliably and is annoying and time consuming. -Otto

**A:** Sorry for the inconvenience. Did you check the log file ("output.log") on Puhti Web Interface? This would sometimes help you diagnos the issue. It is possible that required memory could have reached more than allocated. Your job may have been killed and consequently notebook is disconnected from batch job.

- **Q145** How do we load the numpy module in puhti? Or is there any numpy module in puhti?-Herlik

**A:** numpy has been installed as a part of some modules such as python-data. you can try : ``` module load python-data```. This module has a lot more python packages.

- **Q146** How to config the Rahti for a course - Haining
**A:** Could you please elaborate bit on the scope of course ? We have few more easy options for setting up courses (like CSC notebooks and Puhti web interface) at CSC.
**Q:** It's the bigdata platforms course (link of last year: https://big-data-platforms-22.mooc.fi/). We used ngshare to release and grade assignments with students. It used work before last year, but we met the glustref storage mounting issue on Rahti and it turned to be unuseable for students. I think I'll need some help of setting it up this year. I will check with Rahti experts here.

Just curious, can you containerise your course environment in Jupyter notebook? In that case, you can also use CSC notebooks.

- **Q147** I have created a container in puhti, it handles batch jobs well but when i try to run it in jupyter notebook in the same environment. The libraries are not working despite linking to the virtual environment. what could be the issue. (ian)
- **A:** You can use the "Custom path" option for tykky environments, but then it needs to have everything included, including jupyterlab package.

Is the Jupyter notebook already installed as part of the container?
- **Q148** I would like help with creating allas bucket! - Kim
- **A:** Super, Kimmo will explain this and as a bonus, some encryption tips in room 2!

## 2023-06-21 session

Short talk: No short talk today
(HPC crash course moved to fall due to sickness)

### Zoom breakout rooms and topics

Room 1 - Pouta questions
Room 2 - Machine learning
Room 3 - terminology
Room 4 -
Room 5 -
Room 6 -

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q139** OpenMP and offloading, Touko Puro. When using OpenMP offloading with Fortran is it possible to use `!$omp` workload in offloaded regions? I have functions that I want to offload with OpenMP which use the Fortran array operations which are implicit loops. They are possibly to multithread on CPUs without unrolling the loops with `!$omp` workload, with `!$omp` target and target regions is there any way to spread the work to multiple threads other than unrolling the loop and using `!$omp` distribute teams do?
- please contact [e-mail]
- **Q140** How can I prevent my virtual machine from being deleted by other project members in cPouta? (Yumo)
- A: Hello Yumo, you can lock your instance to prevent accidental deletion. You can use the web interface or use the command `openstack server lock`. Keep in mind that if you want to add a Floating IP, edit Security Groups, attach a volume... you need to unlock your instance, do the change and lock it back.
- **Q141** I am currently writing my report and I got a bit confused with some terminologies used on CSC. What are the differences between the processors, cores, nodes, tasks, and processes. (Tee)
- A: Maybe some helpful pictures on these slides: https://a3s.fi/CSC_training/02_environment.html#/brief-introduction-to-hpc-environments
- The link is not displaying any slides. Seems like just a front page of the material. -> this is a slide set, use the arrow keys or page up/page down to move between the slides!
- **Q142** Using GPU's on Puhti. Trying to utilize GPU's on Puhti while trying to do Quantum machine learning with a container. Frans P
- A: You just need to have the CUDA libraries in the container, and use `--nv` flag when running container. We'll get back to the issue over email later if needed.
- **Q143** I am having some issues with the versions of openfoam on CSC. I asked earlier, and I have tried out the suggestion, but it still doesn't work. Could I get someone to look at the case? Tee
- Unfortunately our OpenFoam specialist is on his summer vacation, but please contact [e-mail] . Esko will reply when he returns!

## 2023-06-07 session

Short talk: No short talk today

- **Q137** Would that be possible to extend max memory (RAM) in Puhti interactive session? Currently it is 60GB, which is limitating for data analysis tasks. Dimitri Meistermann (Univ. of Helsinki).
- It is possible to run interactive jobs with higher memory: just start the session and wait (goes to a different partition).

- **Q138** Question on Comsol crashing, no error message produced and just waits for the crashed process to finish.
- Not sure if this is a Puhti problem or a Comsol problem. Directed to [e-mail] / Sampo

## 2023-05-24 session

Short talk: [GPU programming course](https://enccs.se/events/2023-06-gpu-programming/)

### Questions

:::danger
If you wish you can type your question here before the session. We'll respond to them during the session. **Please join the Zoom call if you have a question, if you cannot join the call you can send the question by email to <[e-mail]>.**
:::

- **Q127** I noticed that Puhti interactive jupyter sessions appear to be badly oversubscribed, so a simple request (e.g. 4 hours, 4 cores, 16 GB, no GPU) ends up queuing for several hours. In my experience, the situation got severely worse over the past month or so, and it runs counter to the idea of being "interactive". I suspect it could also lead to extra load, with sessions getting unused because they start at a random time when the user is no longer active, or users requesting longer sessions hours in advance of actual need, just to have a chance to be able to work when they need them. Are there plans to upgrade the system in the near future, and is there any chance to at least see the queue and get an estimate of the wait time, so other work can be planned accordingly, or possibly "backfill" (e.g. being able to make shorter requests when the qeueue is full)? I cannot attend at the 2pm slot today, but I'll check here for a typed answer. Many thanks already! -- Till Sawala (University of Helsinki)
- A: Thank you for your question -this document is intended for the questions that are to be discussed in the weekly session. We could discuss this in next weeks session, if you are available then? Otherwise, [e-mail] is there for you :) That said, it seems the memory is fully in use in the interactive queue, we'll try to monitor this better in the future. In the meanwhile, you can try to use the `small` partition instead of the `interactive` if you cannot get it running.
- A: (24.5.2023) the interactive partition was increased with 2 M-nodes and 2 M-IO nodes to meet the increased demand

- **Q130** ePouta, how to get started? (Q from last week) Toni V
- A: Discussed with Kimmo M in room 3
- A: Better instructions for "getting started with ePouta" needed! WHO to contact, from home organisation? Is it always a ticket to our servicedesk first?

- **Q131** We are planning to use cPouta for a course where each student will have their own VM to do the course assignments and need to access their VM from their local machine. The floating IPs seem to be scarce. Our questions are 1. Is there a limitation of the number of floating IPs assigned to a project? 2. Or is there some walkground by which one floating IP can be shared across multiple VMs? 3. The assignments can be memory-intensive, is there a limitation of RAM which one project can apply for? -- Yumo (University of Helsinki)

- A: To answer your questions:
1. After creating your project, you can request more floating IP and other resources if needed. You can send an email to [e-mail]
2. You can't use one floating IP for several instances at the same time. You can set a jumphost, that means one host with one floating IP and then access the other instances through this host. We don't have CSC documentation about this but you can find more information on internet on how to set up a jumphost.
3. Like the first question, you can ask for more resources. Just send an email to this address: [e-mail]

- **Q132** A question about queue optimisation. - Jalmari Tuominen (Tampere University)
- A: Might be fair share issue (user running a lot of jobs will get lower priority), suggested to try Mahti or LUMI. LUMI has lots of GPUs, but will take some time to get the project.

- **Q133** Some issues submiting jobs to slurm from mahti with pip containers. Najwa Laabid (Aalto University)
- A: Using non-tykky Python seemed to work, also try reserving NVME in the Slurm job.

- **Q134** I have registed the GPU workshop one week ago from Helsinki, but still not yet approved. It's still in waiting.
- The main coordinator is Thor from ENCC, `[e-mail]`. Please drop him a message.
- Ok, then I send email to Thor, Thx.

- **Q135** SD desktop follow-up question, regarding R add library, Anna-Leena V
- A: Room 4 -> Room 3, waiting for Kimmo M :)

- **Q136** How can I run Matlab Simulink -simulations on Puhti? / Jyrki S.
- Matlab can be ran with GUI or without. With GUI the easisest way is by openon demand.
- Sampo S. to reply in Rocket Chat!

## 2023-05-17 session

No short talk.

### Questions

- **Q127** I noticed that Puhti interactive jupyter sessions appear to be badly oversubscribed, so a simple request (e.g. 4 hours, 4 cores, 16 GB, no GPU) ends up queuing for several hours. In my experience, the situation got severely worse over the past month or so, and it runs counter to the idea of being "interactive". I suspect it could also lead to extra load, with sessions getting unused because they start at a random time when the user is no longer active, or users requesting longer sessions hours in advance of actual need, just to have a chance to be able to work when they need them. Are there plans to upgrade the system in the near future, and is there any chance to at least see the queue and get an estimate of the wait time, so other work can be planned accordingly, or possibly "backfill" (e.g. being able to make shorter requests when the qeueue is full)? I cannot attend at the 2pm slot today, but I'll check here for a typed answer. Many thanks already! -- Till Sawala (University of Helsinki)
- A: Thank you for your question -this document is intended for the questions that are to be discussed in the weekly session. We could discuss this in next weeks session, if you are available then? Otherwise, [e-mail] is there for you :) That said, it seems the memory is fully in use in the interactive queue, we'll try to monitor this better in the future. In the meanwhile, you can try to use the `small` partition instead of the `interactive` if you cannot get it running.

- **Q128** A question about parallelization with DDP using PyTorch Lightning
- A: Mats in room 1
- A: There is a known problem with the [PyTorch Lightning DDP examples](https://github.com/CSCfi/pytorch-ddp-examples) that it uses only one GPU when using `pytorch/1.13` or `pytorch/2.0`. We're working on fixing this ASAP. As a work-around you can use `pytorch/1.12` for now.

- **Q129** I intend to use object detection algorithms in batch (puhti) which requires to first create a virtual environment. I have created one in my project folder, activated it, but when i seek to install required packages using pip-containerize new --prefix <my_env> requirements.txt i get the error Set CW_DEBUG_KEEP_FILES env variable to keep build files. secondly, Do i have to create a container for a virtual environment?
- A: please join the session now :) This document is only intended for these weekly sessions! Zoom URL: https://cscfi.zoom.us/j/65059161807
- Ari-Matti in room 2
- In this case the problem was that the package needed a newer Python version so that all the dependencies could be statisfied.

## 2023-05-10 session

Short talk: How to use the Julia programming language on CSC's Puhti and Mahti clusters

- https://docs.csc.fi/apps/julia/
- https://docs.csc.fi/support/tutorials/julia/

### Questions

- **Q126** I have a question regarding data export from SD Desktop that I'd like to discuss with the helpdesk -Helka

## 2023-05-03 session

Short talk: Monitoring batch jobs on supercomputer

Link to the materials: https://github.com/jlento/atm-doc/tree/master/slurm

### Questions

- **Q124 Sheng: I have installed GAMS (a software similar to MATLAB or others) in the folder `/projappl/project_xxxx` and add the PATH to `.bashrc`. After using `source ~/.bashrc`, I can import `gams` in the login terminal. But after I relogin, I can not use the command `gams` again. Do you have any suggestion? Thanks in advance!**

- In this case the problem was that file `.bash_profile` was missing so `.bashrc` did not get sourced automatically in login shells. It can be copied from `cp /etc/skel/.bash_profile ~`

- **Q125 Ona: How can I use 4 GPUs on LUMI?**
- please ask from LUMI support (zoom sessions every last Wed of month) or via https://www.lumi-supercomputer.eu/user-support/
- Machine Learning guide may also have information on this topic: https://docs.csc.fi/support/tutorials/ml-guide/

## 2023-04-26 session

No short talk today.

### Questions

- **Q122** Question about running a binary that was compiled for Ubuntu 20.04 and this needs *older* glibc than is available in Puhti.
- Can be solved with Apptainer containers. First build a container, pulling a base Ubuntu 20.04 image from DockerHub (needs to be done only once)

    apptainer build ubuntu20_04.sif docker://ubuntu:20.04

Now you can run your binary like:

    export SING_IMAGE=ubuntu20_04.sif
    apptainer_wrapper exec ./your_binary_file

## 2023-04-12 session

Short talk: how our service desk works and how to write good tickets 

### Questions
  
- **Q120** Doing deep learning on Puhti with the Keras-Yolo3 library (Mats already gave instructions on email) + Atte A

- **Q121** Continuation to my last week's question about using the Vulkan API on Puhti GPU nodes: I noticed that graphical applications using the Vulkan API (https://www.vulkan.org/) such as the `vkcube` example from the Vulkan SDK work through X11 forwarding _if my job gets the first GPU in a node._ I wonder if this could even be a driver bug? In addition, inside the job the GPU I get shows up in applications, e.g. `nvidia-smi`, with index 0 even if the GPU is not the first GPU on the node (GPU index is larger than zero when checking with `scontrol show job -d JOBID`). (ET)


## 2023-04-5 session

Short talk: Mini course for getting started in CSC services 
- Prerequisite mini-course e-Lena page: https://e-learn.csc.fi/course/view.php?id=75


- **Q117** Should Vulkan work on Puhti GPU nodes? At least for me, even a simple example (`vkcube` through X11 forwarding, with SSH compression) does not launch on most nodes. I have found that at least one of the `gputest` nodes (`r02g01`) works, but most GPUs I have got on the `gpu` partition do not work (examples: `r15g07`, `r02g04`). The `vkcube` program selects GPU 0 (Tesla V100), tells that it `"Can't find our preferred formats... Falling back to first exposed format. Rendering may be incorrect."` and segfaults. The usecase is that I have a custom visualization script that uses the Vulkan-based Datoviz library (https://datoviz.org/) for interactively visualizing huge files that don't fit into the VRAM of my workstation's GPU. (ET)

- **Q118** Tykky installation gives the following error: 
[ ERROR ] Installation failed 
[ ERROR ] Set CW_DEBUG_KEEP_FILES env variable to keep build files
How do I deal with it? I was using Mamba when I got it, can't check without it because installation is too slow. (Anna Dmitrieva)

**A**: did you use the following command for Mamba installation:

conda-containerize new --mamba --prefix folder_for_installation yamlfile

 PS: Usually some issues such as you have reported may go away if you do fresh installation with tykky. (```module --force purge``` would be a good idea before starting)
    * Could you copy a little bit longer part of the error message?


- **Q119** I have a problem to log in to e-learn for mini-course e-Lena page since I am not a student but a foreign collaborator / P. Matavulj
    **A:** If you have Haka or Virtu accounts, so account to a Finnish research institute or university, those should work. If not, just email us ([e-mail]) and we can create you account for e-lena.

- **Q120** Question: I was tasked to ask, when I logged in to Puhti and enter my project files, there are several files with a name allas-cli-tools-(random numers and letters) they are some sort of residuals when using allas. What are those and can they be deleted? MK
**A:** Yep, ok to delete. They are left sometimes after running some Allas commands, when there's some hick-ups.

## 2023-03-29 session

Short talk: SSH keys and login safety

- **Q116** I want to create an environent in Puhti and want to install particular packages. (sv)
    - (For Python) In many cases you can load our module and add packages with `pip install --user`, see: https://docs.csc.fi/apps/python/#installing-python-packages-to-existing-modules
    - It might also make sense to see what you have installed before in your home dir (as that can sometimes cause conflicts, or be confusing), it's a bit hidden: `ls ~/.local/lib/share/python3.*/site-packages/`

## 2023-03-21 session

No short talk.

- **Q112** What solution do you suggest for interactive coding with some resource-heavy tasks? I'm mostly working with R. Should the code be first developed with an interactive session with toy data and then re-run with the actual data as a batch job? Any other ways to do this or can you suggest best practices? -Matti Ruuskanen, UTU
    - Yes, this is exactly what we would recommend: use the interactive sessions for developing scripts, making sure they work and running smaller test jobs. Then for larger R jobs that need more resources move to batch jobs. A useful addition to some R batch job can be the use of fast local storage (used as default in interactive jobs): https://docs.csc.fi/apps/r-env/#using-fast-local-storage
    - Also discussion on containers and the best way to have a complicated environment that can be modified/transferred easily

- **Q113** Choosing the parameters for batch job script for training a machine learning model on Puhti
I am training a machine learning model with tensorflow/keras on Puhti and have been able to complete some test runs but would like to confirm about some of the parameters in sbatch file before running the full job. - Patrik Lauha, University of Helsinki

    - General points: use `seff` command to check what your job used. `--cpus-per-task` means how many CPU cores you reserve. When using data loading framework, like `tf.data` in TensorFlow, it can utilize multiple cores to feed data to the GPU, keeping GPU utilization high. Use NVME for datasets with many small files (or TFRecords).

- **Q114** How to run array jobs with array elements that run python script parallelized with dask
    - Specifying `--ntasks=1 --cpus-per-task=40` needed to get the elements running in parallel as well. Python generally cannot use MPI (not sure whether dask could enable this similar to mpi4py). User's previous calculation not utilizing array jobs used `--cpus-per-task=40`, however, so the threading solution should be fine in this case. Note that array jobs is not the best solution if you intend to run 100+ elements, it will cause quite a bit of load on Slurm due to job steps. Here a better solution would be HyperQueue, https://docs.csc.fi/apps/hyperqueue/

- **Q115** Reading in 9 GB database files (fasta files) in an array job
    - Use NVME nodes and fast local drive ($LOCAL_SCRATCH)

## 2023-03-15 session

### Advertisements

- Using CSC Computing environment free online course 12.-14.4.: [event page](https://ssl.eventilla.com/computingenvironment-2023-april)
- LUMI AMD GPU hackathon 17.4.-21.4.: [event page](https://ssl.eventilla.com/event/yWdEZ)
- The CSC Summer School in High-Performance Computing 2023, 27.6.-6.7.: https://ssl.eventilla.com/summerschool2023
- Advanced cloud courses Spring 2023:
    - Rahti Cloud containers advanced course: How to make container images with Rahti, TBD (9:00-12:00). https://ssl.eventilla.com/event/GGAOz
    - Rahti Cloud containers advanced course: Deploying and Using Jupyter Platform on Rahti, TBD (9:00-12:00). https://ssl.eventilla.com/event/zAk60

### Short talk: R/python-based web tools at CSC for reproducible research and teaching

[slides](https://a3s.fi/csc-training/weekly_customer_csc.html)

[CSC Notebooks](https://notebooks.rahtiapp.fi/welcome)
[Puhti Webinterface](https://www.puhti.csc.fi/)
[Rahti](https://rahti.csc.fi/)

### Room and topic

Room 1 - CSC Notebooks (Katri)
Room 2 - Puhti webinterface (Sebastian, Henrik, Robin)
Room 3 - Rahti (Alvaro, Jemal)
Room 4 - General questions about the short-talk (Laxman)
Room 5 - R (Q111) (Heli) -> moved to teams
Room 6 - geoapplication in cloud (Q110) (Kylli)
Room 7 -
Room 8 -
Room 9 -

### Questions

:::danger
Note that if you are typing a question in here, we expect you to join the Zoom session :smiley: Help us and type your **name/initials** on the question!
Otherwise, it's probably best to send a support request to <[e-mail]>! 
This is **not replacing our service desk and ticketing system**!
:::
  

- **Q110** Tatu Leppämäki: *Getting a geocoding service running: should I choose Pouta or Rahti?*

I'm trying to get a web service that'll get coordinates from plain text queries running. [Pelias](https://pelias.io/) seems like a good option, and they have [well documented instructions](https://github.com/pelias/docker) on starting up using docker(files?). 

Since dockers are in play, Rahti seems like the correct service to use, but as someone who hasn't used Rahti or dockers before I'm at a loss on how to get started. Pelias provides a script that sets up the system, but I don't think I can run those commands in Rahti(?). E.g. git cloning and making directories. I'll also need some tens of gigabytes of storage for the database. Sounds like something that could be done in Pouta, but can I then use the geocoder as a web service later on, I wonder. 

So my question is: does this seem like something for Pouta or Rahti? How should I move forward from here?

Kylli: I think these guideline suite better to cPouta. You should set up a cPouta VM of suitable size and then install docker and docker-compose as mentioned in the documentation and then follow the instructions. You can get extra storage to cPouta with volumes, ten of gigabytes should not be any problem.

-> Room 6 (after short talk)

And a related question: I missed the basic cloud courses that were running last week. Are the materials for those courses publicly available?
-> https://cloud-solutions.a3s.fi/index.html

- **Q111** Miika Mäki: I have a couple of things that work on an interactive R-env in Puhti, but not when using batch scripts. The questions are quite specific so a breakout room would probably be most convenient for discussing this:
--**1.** futuremice command for parallel multiple imputation works with the parameters future.plan="sequential" & future.plan="multisession", but not with future.plan="cluster". I wonder whether I have set the MPI clusters correctly together with future(mice). Regular future_map-command works just fine and futuremice should use future in the background for the parallalization.
--**2.** Using a custom font works on interactive session but not in a batch job. Interestlingly, if I run another R file with the exact same settings file, there are no font issues.

    Both of the issues could be due the fact that the R-env on Puhti is outdated, but could also have to do with the setup. This would probably be fastest over zoom rather than sending emails back and forth.

-> Room 5 (after the short talk) [Roger]
 Discussed about finding the best way to parallelize the analysis, suggestion to send test data to [e-mail] . No easy solution for the font problem either, could be some small thing going wrong in one script but not in the other?


## 2023-04-5 session


### Advertisements

- The CSC Summer School in High-Performance Computing 2023, 27.6.-6.7.: https://ssl.eventilla.com/summerschool2023
- Advanced cloud courses Spring 2023:
    - Rahti Cloud containers advanced course: How to make container images with Rahti, 24.4. (9:00-12:00). https://ssl.eventilla.com/event/GGAOz
    - Rahti Cloud containers advanced course: Deploying and Using Jupyter Platform on Rahti, 11.4. (9:00-12:00). https://ssl.eventilla.com/event/zAk60


### Room and topic

Room 1 - Pre-course support session for "CSC Computing Environment" course (Rasmus, Laxman)
Room 2 - Tykky problem (Henrik)
Room 3 - 
Room 4 - 

### Questions

:::danger
Note that if you are typing a question in here, we expect you to join the Zoom session :smiley: Help us and type your **name/initials** on the question!
Otherwise, it's probably best to send a support request to <[e-mail]>! 
This is **not replacing our service desk and ticketing system**!
:::
  

- **Q117** Should Vulkan work on Puhti GPU nodes? At least for me, even a simple example (`vkcube` through X11 forwarding, with SSH compression) does not launch on most nodes. I have found that at least one of the `gputest` nodes (`r02g01`) works, but most GPUs I have got on the `gpu` partition do not work (examples: `r15g07`, `r02g04`). The `vkcube` program selects GPU 0 (Tesla V100), tells that it `"Can't find our preferred formats... Falling back to first exposed format. Rendering may be incorrect."` and segfaults. The usecase is that I have a custom visualization script that uses the Vulkan-based Datoviz library (https://datoviz.org/) for interactively visualizing huge files that don't fit into the VRAM of my workstation's GPU. (ET)

- **Q118** Tykky installation gives the following error: 
[ ERROR ] Installation failed 
[ ERROR ] Set CW_DEBUG_KEEP_FILES env variable to keep build files
How do I deal with it? I was using Mamba when I got it, can't check without it because installation is too slow. (Anna Dmitrieva)

**A**: did you use the following command for Mamba installation:

conda-containerize new --mamba --prefix folder_for_installation yamlfile

 PS: Usually some issues such as you have reported may go away if you do fresh installation with tykky. (```module --force purge``` would be a good idea before starting)
    * Could you copy a little bit longer part of the error message?

- **Q119** I have a problem to log in to e-learn for mini-course e-Lena page since I am not a student but a foreign collaborator / P. Matavulj
    **A:** If you have Haka or Virtu accounts, so account to a Finnish research institute or university, those should work. If not, just email us ([e-mail]) and we can create you account for e-lena.

- **Q120** Question: I was tasked to ask, when I logged in to Puhti and enter my project files, there are several files with a name allas-cli-tools-(random numers and letters) they are some sort of residuals when using allas. What are those and can they be deleted? MK
**A:** Yep, ok to delete. They are left sometimes after running some Allas commands, when there's some hick-ups.


## 2023-04-12 session

Short talk: how our service desk works and how to write good tickets 

### Advertisements

- **Practical Deep Learning course**, 3.5.-5.5.: https://ssl.eventilla.com/event/8aPek
- The CSC Summer School in High-Performance Computing 2023, 27.6.-6.7.: https://ssl.eventilla.com/summerschool2023
- Advanced cloud courses Spring 2023:
    - Rahti Cloud containers advanced course: How to make container images with Rahti, 24.4. (9:00-12:00). https://ssl.eventilla.com/event/GGAOz


### Room and topic

Room 1 - 
Room 2 - 
Room 3 - 
Room 4 - 

### Questions

:::danger
Note that if you are typing a question in here, we expect you to join the Zoom session :smiley: Help us and type your **name/initials** on the question!
Otherwise, it's probably best to send a support request to <[e-mail]>! 
This is **not replacing our service desk and ticketing system**!
:::
  
- **Q120** Doing deep learning on Puhti with the Keras-Yolo3 library (Mats already gave instructions on email) + Atte A

- **Q121** Continuation to my last week's question about using the Vulkan API on Puhti GPU nodes: I noticed that graphical applications using the Vulkan API (https://www.vulkan.org/) such as the `vkcube` example from the Vulkan SDK work through X11 forwarding _if my job gets the first GPU in a node._ I wonder if this could even be a driver bug? In addition, inside the job the GPU I get shows up in applications, e.g. `nvidia-smi`, with index 0 even if the GPU is not the first GPU on the node (GPU index is larger than zero when checking with `scontrol show job -d JOBID`). (ET)