# IO-SEA Review
###### tags: `EUROHPC`
2 platforms
IT4I platform and the DEEP test cluster
Ready to be benchmarked on DEEP cluster (2nd half of the project)
M24?
Co-design to develop/find out what is required from users.
Pre-fetching to data node: Any collaboration with other EuroHPC projects?
Workflow Description File.
Any discussions with other projects?
Benchmark infrastructure (shared with DEEP-SEA?)
Weekly scheduled to run benchmark.
## TSMP
different executable (ESM)
Run on DEEP prototype.
Hypothetical benchmark result (based on what? Simulation tool?)
IT4I platform for developing technologies. Small-scale workflows. Can you run applications?
Size for applications? Applications will be integrated in JUBE so they can run on several platform.
Run Darshan IO traces (for ADMIRE).
## WP2 Data workflow / Ephemeral Data Access Environment
Motr/S3 VM up and running at IT4I
no applications can use S3 servers directly.
or through DASI (and DASI gateway)
SBF/GBF temporary
Slide 59
T2.1 Ephemeral Data Access Environment
T2.2 NVMe and Non Volatile Memory usage
NVMe can be used as a standard disk
Non-volatile Memory --> peristent storage
Compute Express Link
Optane cancel (Intel).
Impact on the project?
Rethink this task.
Simulator for CXL Technolofy to be used (maybe)?
T2.3
Slurm plugins development : any collaboration with
other euroHPC
Enhance ParaStation
## WP3 Instrumentation & Monitoring
Wrokflow level to get instrumentation (and not job level) Implemented in IO instrumentation and LLview.
4 to 5 use cases but now all 5 are included.
3.1 Instrumentation and design
slide 80 --> workflow view instrumentation
do you need workflow.yaml?
IO instrumentation: made changes to IO instrumentation to get a better overview of the applications.
Do you have enough use case to have a "generic" solution?
Slide 84: LLview is always by job and not by workflow --> aggregated by workflows.
Prepare workflow report for who?
Implement ephemeral services too (instrumentation)
overhead?
T3.3
T3.4 AI based analytics
Design of the solution is done.
model
- hierarchical storage performance model
user selects performance or energy-efficiency strategy.
Is it based on 1 run only?
Does it take into account past runs?
Data movement log database? What is it?
How does a user chose between the different storage e.g. how does user write workflow file?
## WP4 Hierarchical Storage Management Features
Achievement parallelisation of Phobos
Redirect --> any overhead?
Slide 107 --> How many IO requests do you take into account?
When you group operations?
+ Grouping I/O on tapes Slide 108
How many IO do you group?
Slide 109: rpacking of sparse tapes
to reuse tapes.
RAID algorithm but for tapes.
T4.2 Vertical data placement policy.
Alide 111: MongoDB One data base for all
Or does it scale?
Collect info about objects.
T4.3 HSM Interfaces/HSM API (Hestia)
Public API exposed to DASI/S3/POSIX
4.4 Migration Paths.
5.4 EUPEX task to integrate IO-SEA component in EUPEX.
IO-SEA will be over when EUPEX Pilot will be up and running.
## WP5 Application interfaces
Task 5.2 POSIX for legacy support
--> to be deployed as an ephemeral service
Parallel to DASI for use cases which can't easily modify their IO stack.
Weather --> using POSIX layer.
Target audience?
Who is doing what?
malleability? ADMIRE?
Slide 146: integrate POSIX interface for use-cases which require it.
In-situ processing and visualization.
slide 148.
WP6: Dissemination & Outreach.
DASI interface development?
Support? Beyond the life time of the project? Sustainability?
Standardization Connect with DEEP-SEA for MPI.