All Members Community

 View Only
  • 1.  Computer science research computing core

    Posted 5 days ago

    Hi Core geniuses! 

    We are digging into our research computing core and am hoping to get some different options for billing. Do any of you have a research computing core and if so how are you billing users? The current model we are using is not sustainable so i am hoping that you guys have great models that we can recreate at UNC.

    Any information you have would be greatly appreciated.

    Meghan



    ------------------------------
    Meghan Kraft
    Director of Research Core Facilities
    University of North Carolina
    Chapel Hill NC
    ------------------------------


  • 2.  RE: Computer science research computing core

    Posted 2 days ago

    Hi Megan, 

    This is a really interesting topic and timely as we're looking at this too.  In our case our research software, data storage and processing infrastructure and support at embeded in the corporate digital solutions arm of the university and there hasn't been any cost recovery to date. 

    The first task for us is to tease out what should be institutionally funded and what should be cost recovered.  There are also more granular questions for data storage around what should every researcher be provided by the university, and the inevitable inequity for disciples that created significant data compared to those that don't.

    I'm really interested to hear how other institutions run their digital research support cores.



    ------------------------------
    Dr Sarah McAtamney
    Senior Research Infrastructure Officer
    Griffith University
    Queensland, Australia
    ------------------------------



  • 3.  RE: Computer science research computing core

    Posted 2 days ago
    Hi.

    Here are the CU Boulder rates: https://www.colorado.edu/rc/resources

    We have a mix of free and fee-based structures.

    Best,

    Joe

    -----

     

    Joe Dragavon
     
    Director, Core Facilities and Shared Instrumentation

       Research and Innovation Office
    Director, Advanced Light Microscopy Core 

       BioFrontiers Institute

    JSCBB C315

    Teams/Office: 303.735.6988
    Cell: 720.934.2933
    University of Colorado Boulder
     

    Arrange a Consultation or a Meeting: 43ec276be94b4ad5bdb7519f84eda1dc@colorado.edu?anonymous&ep=plink" title="https://outlook.office.com/bookwithme/user/43ec276be94b4ad5bdb7519f84eda1dc@colorado.edu?anonymous&ep=plink" data-outlook-id="3adf8a36-9a41-4acd-bb21-3d131e2089ba" style="color: #3C9AFF; margin-top: 0; margin-bottom: 0">Click Here

    Website: https://advancedimaging.colorado.edu/
    Need an Instrument?: https://www.colorado.edu/sharedinstrumentation/core-facilities-collections 

     






  • 4.  RE: Computer science research computing core

    Posted 2 days ago
    Meghan,

    At Florida State, we have a Research Computing Center (RCC) that is effectively a core facility although not necessarily labeled as such. In my opinion, the biggest mistake they made in their funding model was having to recover staff salaries as part of the charges to researchers. Even with cost sharing from Vice President for Research, which is typically a 100% match,  the cost of the resources (CPU, GPU, and storage) is probably at least 20-30% above what you can get quoted from vendors. Admittedly, some of that isn't a direct comparison, like 5x 20 TB in a RAID6 vs. 60 TB in GPFS on the cluster's low latency SAN, but it can be difficult to convince the less technical PIs that the cost increase is justifiable.

    On top of that, since PIs don't own the hardware (they buy "computing units" or "storage units") and it's not cost effective for the University to run the aggregate amount of hardware off warranty, the capacity has to be repurchased after 5 years when the extended warranty expires. That creates a lot of fiscal pressure on grant funded research in comparison to buying their own and running it until it dies. It also annoys the faculty that the off-warranty hardware gets repurposed within the RCC but is no longer dedicated to their use. That leaves them feeling that they paid more than full price, due to the cost overhead, while only getting 2/3 of the lifetime usage. Consequently, departments that have the expertise and infrastructure (like ours) tend to do it themselves.

    I can elaborate, either on or off list if there is interest, but didn't want to hit everyone with a huge wall of text on Monday morning.


    Mike

    -- 

    Michael Zawrotny, PhD.

    Director of Computer Resources

    Institute of Molecular Biophysics

    91 Chieftan Way

    Florida State University

    Tallahassee, FL 32306-4380

    email:  mzawrotny@fsu.edu

    phone:  (850) 644-0069

     

     






  • 5.  RE: Computer science research computing core

    Posted 2 days ago

    Hi Michael,

    Thanks for following up. Is your RCC available to all campus or just to a department? Right now this core i am talking about is about 99% used by Computer Science and is run out of their department. In the past they have been lucky to get funds to purchase new nodes but those days are coming to an end so they have had to build in depreciation. They also are currently being subsidized by the College of Arts and Sciences and they want to reduce that subsidy over the next year or 2 to almost nothing.

    Since this core is run and mostly utilized by CS and not broadly on campus we are trying to find a more sustainable rate model as right now they are using a standard monthly rate they have calculated and then only charging it to grants that the faculty has direct effort on. 

    I hope that provides a little more clarity.

    Happy to get on a call too!

    Meghan



    ------------------------------
    Meghan Kraft
    Director of Research Core Facilities
    University of North Carolina
    Chapel Hill NC
    ------------------------------



  • 6.  RE: Computer science research computing core

    Posted 2 days ago
    Meghan,

    I apparently misunderstood what you were asking about. Yes, the RCC here is a university wide center, so completely different than your situation. The first section below discusses starting a new, centrally managed compute core; the second section is about keeping it at the department level. There is some redundancy in the two that I left in for context.

    Judging by the job title in your signature, it looks to me like you're considering lifting the CS run compute up to a university-wide center. If that's the case, and the reason to eliminate the A&S subsidy, then I think the first consideration would be whether or not your equivalent to our Vice President for Research would underwrite some or all of that. If the CS department has been putting in money to run the current system, a similar sized, periodically updated direct charge to CS might be feasible. If the compute facility is going to become a university resource, partnering with other Colleges, e.g., Engineering or Medicine might also be viable. Admittedly, the latter could be problematic if any patient information were to be involved, as that would require HIPAA compliance.

    Once you've exhausted those avenues, I think you're stuck with the traditional approach of charges to grants for compute time and storage. That's a pretty tough problem because you can easily set rates that are either too high for the PIs to be willing to pay, or as has happened at least once here, high enough that a granting agency's program officer vetoes that item in a grant proposal. You'd have to look at the rates around the country to ensure that yours are not too far above that. Those charges can be built into future grant proposals, but you'd need some kind of institutional support for existing grants that lack that.



    I think that what your current situation is more like what I run for our Institute. We do have a core auxiliary that covers cell culture, protein purification, and our instrument lab, but my facility was deliberately left out of that. All of our facility directors have non-tenure earning appointments via the College, so there is no need to recover salaries through charges, and the way we run the computational facility is designed not to require charging grants (see below).

    My facility has a mixture of GPU compute nodes, workstations (with and without GPUs), and a bunch of file servers and NAS boxes. The common infrastructure, mostly home directories and authentication, is department funded, and the items dedicated to a given research group is bought from that group's grants. We used to have some Institute purchased compute nodes, but those went away at least 15 years ago.

    This division is necessary due to the way things are funded here at FSU. Potential revenue streams include: operational budget through the College (Arts & Sciences), per student credit hour kickbacks, department portion of grant indirect costs, and direct grant funding. Our Institute is not a full department, so we don't have any appreciable course credit hours, making that one a big zero. In addition, the department cut of indirect costs is split with the PIs home department (reflecting startup package distribution) and the operational budget cannot be used for capital equipment, which is anything over $5K.

    If we had credit hours, we could utilize some of that to purchase new nodes. Likewise, the University has a student "tech fee" that departments can submit proposals to for projects that benefit student learning in some fashion. The sticky point for either of those would be a mixed-use research and academic cluster. I know for a fact that the tech fee committee takes a really hard look at projects they think could be research posing as instruction in order to get "free money".

    Funding compute is a tough problem. If I had to do it, I'd start off with looking at the available revenue streams like what I describe above, along with the proportion of course related to research computing. Then, if possible, use any appropriate academic funding or staff positions to reduce the operational cost basis. 

    If the goal is to eliminate the A&S subsidy, that starts with a hard look at how much of the total cost comes from that. A 20% subsidy is a lot easier to replace than 70%. If the compute resource is essentially all research, can you replace the A&S portion with one through your equivalent to our Vice President for Research? Does Chapel Hill have any kind of internal grant system you can tap into? We have an internal small instrument grant system that goes up to around $50K (or a bit higher with Department matching commitment). It might be possible to argue that compute is the Computer Science equivalent to scientific instruments.


    Mike

    -- 

    Michael Zawrotny, PhD.

    Director of Computer Resources

    Institute of Molecular Biophysics

    91 Chieftan Way

    Florida State University

    Tallahassee, FL 32306-4380

    email:  mzawrotny@fsu.edu

    phone:  (850) 644-0069