Edu-IT Infrastructure Needs Assessment: Cloud and Server Infrastructure 2021-23

--- tags: eduit --- # Edu-IT Infrastructure Needs Assessment: Cloud and Server Infrastructure 2021-23 # By Kristoffer L. Nielbo and Peter B. Vahlstrup Center for Humanities Computing Aarhus, Faculty of Arts, Aarhus University, Denmark :::info __Executive Summary__ This documents describes Aarhus University Arts' current needs (2021-23) for cloud and server infrastructure with an educational purpose (edu-IT Infrastructure). Three categories of infrastructure needs are identified, _Compute Servers_, _Web Servers_, and _General Services_. The categories target students from Cognitive Science, Information Studies, and Cultural Data Science. The proposed solution is however fully scalable and can meet future needs for Edu-IT Infrastructure. We recommend a solution that use and evaluate existing AU infrastructure (Azure and UCloud) combined with local hosting of services and alternative commercial cloud providers. ::: ## Introduction At Aarhus University Faculty of Arts we are seeing a rapid increase in the needs for cloud and server infrastructure with educational purposes. Broadly speaking, these needs fall in three categories: a) __Compute Servers__, shared access to compute resources for processing-intensive and machine learning applications; and b) __Web Servers__, shared access to hosting resources for developing and running web applications, and c) __General Services__, shared access to contemporary tools for collaboration, development, and version control. In all categories, educators and students need to have support for installation and maintenance of specific software (e.g., JupyterLab, Node.js, GitHub CLI/Desktop). For students the _current situation_ is such that they have to rely on their personal laptops or third-party cloud vendors with all the security risks (e.g., no data management agreement) and reinforcement of prior inequalities (e.g., income level) that entails. For educators there are no standardized solutions; AU-IT can to some extend support set-up of virtual machines within the limits of AU's Microsoft Azure agreement. Azure however comes at a very high price in comparison with previous solutions such as local hosting with virtual private servers at AU-IT or departmental servers. In some cases, educational programs can have old solutions that are still working but with limited functionalities and AU IT plan to phase out dated solutions. As part of Center for Humanities Computing Aarhus' Covid-19 task force, we developed ad hoc edu-IT solutions during spring 2020. These solutions primarily consisted of allocating server space or offering applications for education (e.g., JupyterHub). It soon became apparent that several educational programs, e.g., Cognitive Science (CogSci), Information Studies (IS), Cultural Data Science (CDS), and Linguistics, have a systematic need for edu-IT infrastructure that goes beyond educational software (e.g., learning management systems). Over the last couple of years, the need for such infrastructure has been growing rapidly as almost all research and educational activities have been impacted by digitization and development of computational sciences. Apart from promoting diversity among students and ensuring a high level of information security, seamless access to cloud computing will give students and educators access to _state of the art applications_ that reflect requirements from the global digital workplace. It ensures _scalability_ that makes us capable of responding to shifting working conditions (e.g., Covid-19) and application requirements (e.g., GPU Accelerator for machine learning) at a _competitive price_. Last, but not least, it allows AU to promote _proper data curation and management_ practice to students and educators in accordance with the _FAIR_ principles, Findable, Accessible, Interoperable and Reusable. Following the general movement towards cloud, AU-IT have replaced their earlier Virtual Private Server (VPS) service with the commercial cloud vendor Micrisoft Azure. While this is in accordance with our suggestion of offering access to cloud computing, it is very expensive (4x-25x) when compared to other commercial cloud vendors (e.g., Contabo). Even if this price difference is leveled out, which there is no reason to expect, we still need to provide seamless access for students and educators. This requires both development, maintenance and support resources for years to come. It what follows, we will describe s series of use cases and evaluate needs based on our experiences with these use cases. We want to emphasize that while the cases are specific to a small set of educational programs, the requirements will grow for many Arts' programs following the general digitization trend. ## Active CHCAA Projects ## CHCAA is currently running multiple edu-IT infrastructure projects that provide cloud and server access to students and educators. All projects are undertaken in order to identify and evaluate existing solutions. The projects can be grouped by the infrastructure they use, UCloud or local server, or whether they target students directly or indirectly through educators. ### UCloud for Students ### $\alpha$Cloud is an IT-West funded infrastructure project the makes UCloud available for CogSci master students, in particular two 10 ECTS courses (Natural Langauge Processing and Advanced Cognitive Neuroscience), and the 10 ECTS Data Science course. We allocate 5000 DKK for compute, 500 DKK for storage, and 30hrs support for one admin from CHCAA pr. course. Students use UCloud to analyze data, train models, and solve course exercises. An added benefit of using UCloud is that students get experience with cloud infrastructure, containerized applications, and virtual machines that reflect a modern work environment. Because UCloud is intended for researchers primarily (i.e., DeiC's Type 1 system), extensive educational use can encounter issues with scalability (e.g., few research teams need 30+ people to access one project) and resource allocation (i.e., only the 50\% compute allocated to universities can be used for educational purposes). ### UCloud for educators ### Apart from educator access through $\alpha$Cloud, CHCAA has offered individual course components in UCloud through the competency development course Digital Curriculum. For individual course modules (e.g., image analysis, information visualization), CHCAA offers 15 hrs for development/support/tutorial pr. module. ### Local servers for students ### CDS and master thesis students at CogSci are currently provided access to local servers at CHCAA for processing-intensive applications and simulations. For CDS this amounts to three courses with 35 users each that have access to Jupyter and RStudio (admins included) and for CogSci we have a total of 10 user accounts (three of which require specialized hardware). CDS has been allocated 60hrs support from CHCAA in total. CogSci master thesis students are allocated 3 hrs/user on average. ### Local servers for educators ### Participants on Digital Curriculum and educators at CogSci and CDS are are provided access to local servers by CHCAA for workshops and teaching planning. We currently have 33 users. Digital Curriculum has 0.125 FTE for support, while educators get support as needed, approx. 2-3 hrs/usr. Additionally, CHCAA has two Edu-IT projects CT-Code-Camps and Introduction to digital methods in humanities research that will use a mixture of local servers and UCloud in the fall 2021. ## Requirements and recommendations __Web Servers__ Web servers for courses, course development, and, ideally, for students between course. For Arts specifically, we recommend additional research and development by CHCAA in order to determine scalability, legality, and environmental and economic sustainability for commercial cloud vendors vs. local hosting. To ensure access to web servers during fall 2021/spring 2022, we suggest Azure at 129-220 DKK/month pr. student. Regarding tools on web servers (e.g., node.js, nginx/apache2, mongodb/mariadb, Git), we recommend that CHCAA and AU-IT collaborate in development of base images that provide seamless access to the required software configuration for web development. Images ensure efficiency, agility, and re-usability of applications and resources. __Compute Servers__ To provide access to compute for courses, course development and BA/MA thesis projects, we recommend to use UCloud primarily and local servers secondarily (for specialized hardware). Through UCloud we are already offering a range of applications and CHCAA has developed solutions that ensure persistence between course modules. Continued utilization of UCloud's resources require either that AU's eResource committee acknowledges the need or that Arts buy additional provisioning of compute resources through the Type 1 consortium or a combination. UCloud for education will continue to require support 20-30 hours support pr. course for maintenance and development. In special cases, especially for master thesis students, it will remain a price and security advantage to provide student access to CHCAA servers. These cases typically require limited support 2-3hrs, because formulation of the student projects presupposes skill/knowledge. __General Services__ Finally, we have received several requests from educators and students for cloud-based collaborative authoring tools in markup languages such as Markdown, e.g. Hackmd, and LaTeX, e.g., Overleaf, communication platforms, e.g., Slack or Matrix, and version control in Git with GitHub or GitLab. These tools are commercial and typically provided through software-as-a-service model with a monthly user fee. Some Danish universities, e.g., AAU and DTU, have institutional access to Overleaf through Overleaf Commons. CHCAA has not submitted an Overleaf request for service and therefore has no pricing currently. For several services, e.g., GitLab, Hackmd and Matrix, it is possible to host local instances with a license. This dramatically reduces the cost, for Hackmd and Matrix it is even free, but will require hardware and additional maintenance from AU's side. For version control, AU-IT already has a GitLab local server, but it is not rolled out to students. In order to determine needs in more detail, we suggest additional research and development, specifically for local hosting of Hackmd and Matrix. Apart from reducing price, local hosting also avoids issues relating to third-party data processing agreement. For Overleaf, we would suggest that CHCAA submit an Overleaf request for service to AU-IT for a center account that can be used to evaluate educational needs. __Table 1__: Courses that currently need Educational-IT infrastructure. Table 1 is based on active projects and needs assessment at CogSci, CDS, and Information Studies (INF). We have no reason to expect that the infrastructure need will not increase for these educational programs, nor in other educational programs, e.g., _Scandinavian Languages and Literature_, _History_, _Linguistics_. In table 1, the needs assessments for CogSci and CDS are based on active projects, for Information Studies they are based on interviews with educators. We have not included CogSci Bachelor for lack of estimates, but they will be similar to the Master program. | Course name | Program | Infrastructure | No. Students | | --- | :---: | :---: | :---: | | Natural Language Processing | CogSci, Master | _Compute_ | 40 | | Advanced Cognitive Neuroscience | CogSci, Master | _Compute_ | 40 | | Data Science | CogSci, Master | _Compute_ | 20 | | CDS Introduction | CDS | _Compute_, _GS_ | 30 | | Visual Data Analysis | CDS | _Compute_, _GS_ | 30 | | Textual Data Analysis | CDS | _Compute_, _GS_ | 30 | | Spatial Data Analysis | CDS | _Compute_, _GS_ | 30 | | Data studier | INF, Bachelor | _Compute_ | 110 | | Organisationsetnografi | INF, Bachelor | _Web_, _GS_ | 110 | | Forretnings- og systemudvikling | INF, Master | _Web_ | 25 | | Programmering på internettet: Front-end | INF, Master | _Web_ | 30 | | Programmering på internettet: Back-end | INF, Master | _Web_ | 30 | ## Development and operational costs 2021-23 Implementation of Educational IT infrastructure (i.e., the three categories) requires staff hours either at CHCAA or AU-IT (preferably a combination). Given CHCAA's active projects, we estimate it to 1.5 FTE IT Developer/architect and systems administration for development, testing, and maintenance of all three categories with a particular focus on _web servers_ and _compute_ given the immediacy of the needs. For compute servers, we hope to continue to use Aarhus University's Type 1/UCloud provisioning. The cost for compute and storage is 5,500/course, but for courses within the same educational program (e.g., CDS or CogSci master) it is possible to bundle courses in sub-projects, thereby distributing resources across courses. We estimate that 2,500 course should be sufficient (total: 50,000 DKK). For web servers, autumn 2021-22 with Azure will cost 15,000 DKK/course (total: 75,000-100,000). By 2023, we expect to have a more sustainable solution in place with substantially large web-servers for 8,000 DKK/course (total: 40,000-53,000). Additionally, for 2020-23, we need to purchase hardware for 50,000 DKK to develop and test local hosting of general services and web servers. The hardware will continue to be used for general services to students. :::info __TERMINOLOGY__ __Cloud computing__: refers to on-demand availability of computing power and storage that, in this context, is either provided through DeIC or third-party cloud vendors. __Compute Server__: a server specifically designed to run processing-intensive applications. __Image__: A template for virtual machines or container that contains a software configuration. __Local server__: refers to access to hardware and services (e.g., data sharing and compute resources) hosted either directly by CHCAA or through AU-IT's old VPS service. __Machine learning__: development and applications of algorithms that improve automatically through experience ("learn") and by the use of data. __Processing-intensive application__: applications for analysis or simulation that are too large or complex to be dealt with by traditional applications. __Virtual Private Server__: a virtual machine provided as a service by, in this case, AU-IT. __Web server__: software that can satisfy client HTTP requests on the World Wide Web :::