GPU Resource Management Policy

on the HUN-REN Cloud from 1 January 2025
For IaaS projects

1. Introduction

As of 1 January 2025, a new GPU Resource Management System will be in place for HUN-REN Cloud users for IaaS-type projects using GPU. Introducing the new system aims to enable more GPU-based IaaS projects to be launched on the cloud in a unit of time (e.g. per year).

2. Categories of IaaS-type projects using GPU

IaaS projects using GPUs are divided into 4 categories:

• Projects with a 1-year requirement, which receive the requested resources for a full year
  • Project providing service
  • Umbrella project
• Projects requiring 3 months, which will receive the requested resources for 3 months
  • Preparatory project
  • Advanced project

The 3 months means that projects that cannot request 1 year on the cloud at a time (as they are not providing a service and are not umbrella projects, but individual projects) can only request 3 months of cloud usage at a time. However, if they see that 3 months will not be enough to complete the project, they can request an extension of the project, but this request can also be for 3 months only. However, the 3-month extension request can be submitted as many times as you wish.

2.1. Project providing a service

The 1-year service project will provide an ongoing service to external users (e.g. facial recognition service). Considering that HUN-REN Cloud provides best effort service without SLA, projects providing services running on the cloud can only provide such best effort service to their users and cannot claim SLA from HUN-REN Cloud. The HUN-REN Cloud user requesting a service project must indicate this acknowledgement on the project request form when submitting a project request.

2.2 Umbrella project

A 1-year umbrella project is a project in which not one but several projects/research teams share the resources requested and organise the allocation and scheduling of resources among themselves. For example, projects in a university faculty share the GPU resources allocated to them. When applying for an umbrella project, the project application form should indicate the minimum number of projects that will be part of the umbrella project and should specify, per project, the participants and the number of publications and dissemination activities that will be undertaken.

2.3 Preparatory project

The 3-month preparatory project is a project that is asking for GPU resources on the cloud for the first time and the researchers involved have no experience of using GPUs on the cloud.

Such projects typically work to develop the software environment in which the GPUs will be used. This is usually a time-consuming task and does not involve intensive GPU usage. Therefore, preparatory projects may only require the smallest GPU flavour with 1 GPU.

Filling out the 3 months is optional. If the project is completed before then, you can close the preparatory project and apply for a new advanced project (see advanced project application form).

2.4 Advanced project

A 3-month advanced project is a project that already

• has either successfully developed the software environment needed for the project
• or has a member who has already used HUN-REN Cloud with GPU and has experience in setting up a GPU software environment.

Typically, a preparatory project can become an advanced project after its successful completion. Advanced projects can now require any GPU flavour. Successful completion of a preparatory project must be demonstrated by a presentation to the project that wants to move from preparatory to advanced status.

3. Concepts for deciding project needs

The GPU allocation system can be in one of two states:

• Default state (there is enough GPU capacity to meet the current project demand)
• Competition state (not enough GPU capacity to meet the current project demand)

We distinguish three types of periods:

• 3-month Project periods (implementation period for 3-month projects)
  (Jan 1 - Mar 31), (Apr 1 - Jun 30), (Jul 1 - Sep 30), (Oct 1 - Dec 31)
• Evaluation periods (the period during which pending project applications are evaluated)
  1-14 Mar, 1-14 Jun, 1-14 Sep, 1-10 Dec.
• Preparation/Closure periods (in case of an approved new project, preparation for the next project period, in case of a rejected extension project request, closure of the project)
  15-31 Mar, 15-30 Jun, 15-30 Sep, 11-23 Dec.

We distinguish between two types of project needs

• New project request (the project will be created now on the cloud)
• Project extension request (the project is running on the cloud and wants to extend its operation for an additional 1 year or 3 months according to the project category)

4. New project request fulfilment algorithm

When requesting a new project, the state of the GPU allocation system should be taken into account.

• In the Default state, a new project request comes in and there are enough GPUs, then the new project can be started and the Default state remains.
• In the Default state, a new request comes in that cannot be fulfilled, then we transition to the Competition state and the new project is put on hold.
• In Competition state a new request comes in, the new project will have to wait for the next Evaluation period.

Evaluation of new and extension project requests pending in Competition state:

• If there are enough GPUs for all new project requests (including extension requests), then all new projects can be started and the projects that requested extensions can continue. The system will go to the Default state.
• Not enough GPUs for all new and extension project needs. At this point, projects are evaluated based on the submitted request forms. (New and extension requests are weighted equally.) The system remains in Competition state.

5. Evaluation of project extension requests

Regardless of the status of the GPU Allocation System, you must wait until the next Evaluation Period to have your project extension request assessed. If you wish to continue a running project without stopping it, you must submit a continuation request before the current Project Period's Evaluation Period. If this is done, the project continuation request will automatically be transferred to a proposal, for which the evaluation will be described later.

If the submission of a project continuation request misses the start of the Evaluation Period within the current Project Period, it will have to wait until the next Evaluation Period within the next Project Period to be assessed. Therefore, we consider this project as not wishing to proceed and will give up the resources allocated to it. This is important because these resources will be allocated to other projects. Consequently, the project must move off the cloud before the end of the current Project Period. This is one of the purposes of the Preparation period (preparation for moving the project out of the cloud). In this case, the project must be paused for the next 3 months of the Project Period.

For the above reasons, the project that wishes to apply for an extension must submit its request for continuation in time.

6. Useful explanations

Please note that at the end of the 3-month Project Period, all 3-month projects must either be completed or, if you request an extension in time and the assessment is positive, you can continue without interruption.

This is the reason why during the Evaluation Period, the 3 months of resources allocated to IaaS projects for a given Project Period can be reallocated to projects that are running efficiently.

In a Competition state, all applications received must wait until the next Evaluation Period, when all applications (new and extension project applications) received until then will be evaluated by a panel of experts and the available resources will be allocated to the winning applications.

If the extension and new project requests received by the Evaluation Period can be met with the available resources, the evaluation process will be simplified and all applications will automatically receive a positive evaluation.

As the above shows, if there are enough resources and no competition, then practically everything happens in the same way as in the previous system.

• Those who want to start a project, get permission immediately
• Anyone who wants to extend the project will get permission in time to continue the project without interruption, and because they can request an extension as many times as they like, they can run the project for as long as they like

There are only 3 small changes:

• By implementing the Preparation Project, we avoid GPU resources being idle during the initial software environment setup.
• Every 3 months (or yearly) you will be asked to extend projects, but this will be automatically enabled if there are enough resources.
• For all new and extension project requests, the GPU request form must be filled out (actually only once at project submission, with a minimal update required for extensions at most).

7. Evaluation of project applications and GPU application forms

To be evaluated, all new and renewal project requests must be accompanied by a completed GPU request form on the cloud website. The information provided here will be used by the evaluation committee to assess the projects. It is important to note that new and extension requests are weighted equally (neither is preferred). The completed GPU request form must be attached via the previously used web interface.

The GPU application form requires the following information:

• Which of the four types of IaaS projects does your project require
• A brief description of the aim of the research and its expected scientific impact (max 1500 characters)
• A list of the project members involved in the project who will be actually using the cloud, with their academic classification (PhD, habilitated associate professor, professor, MTA doctorate, academician). Within this, the project leader (who is also the applicant) should be highlighted.
• The expected percentage utilisation of the required GPU resources during the project.
• The expected timing pattern of GPU usage (e.g. continuous, intermittent, how many usage periods and how many breaks between them, etc.) (max 1000 characters).
• The name of international projects or organisations closely related to the project, a description of how they use the results of the project to be launched on the cloud or how they are otherwise related to the project (max 1500 characters).
• Number of publications undertaken (D1, Q1, Q2, presentations, etc.)
• If the Cloud users involved in the project have other running or completed projects using GPUs on the HUN-REN Cloud or other Clouds, an itemized list of the results (publications, presentations, etc.) achieved in the project(s) (in such a case, the preparatory project phase can be skipped and an advanced project can be requested immediately).
• In case of an extension request:
  • Is there a change in resource requirements (reduction or increase)?
  • Results achieved in the current 3-month project period (max 1000 characters).
• Declaration:
  I declare that I will use the project running on the HUN-REN Cloud for the research described in the requested project within the institution specified.

Specialities by project category:

a/ Project providing a service:

Declaration: I acknowledge that HUN-REN Cloud will provide a best effort service without SLA and accordingly the service provided by the requested project will be best effort without SLA. I declare that I do not have and will not have any SLA claim towards HUN-REN Cloud.

b/ Umbrella project:

The minimum number of projects that will operate under the umbrella project and how many of these are advanced projects should be specified.

For those projects that are already known, the title of the project, the participants and the number and type of publications and dissemination activities undertaken should be specified.

c/ Advanced project:

It should be specified why this project could be specified as advanced:

• Successful completion of a preparatory project
• At least one of the project members has already run an advanced project on HUN-REN Cloud

8. Summary

The complete process (application and assessment) is shown in Figure 2.

As long as there is no competition (there are enough resources), everything happens as in the previous system. Once there is competition (resources are exhausted), the start and continuation of all projects are decided in a competitive situation, where the amount of resources a project requires is an important criterion. Therefore, it is in everyone's interest to only ask for the resources they really need and to use them as much as possible. HUN-REN Cloud's team of experts can help you with this, as well as with moving a project from the pre-project phase to the advanced project phase as soon as possible.