I ran some numbers on the core-hour utilization of the big groups in ES and NE ... Total core-hours on Fusion + Blues = 54.8M/year (300*8 + 310*16) * 24 * 365 * 0.85 Sibendu's group has requested 11.14 M core hours from 14 projects Riccardo has requested 1.3 M core-hours from 2 projects Justin has requested 3.1M in 3 projects. Based on these numbers, the EGS directorate has requested close to 28.5% of the available compute time. I have not added other ES & NE projects to this number. Sibendu's group at 11.14M stands at 20% of the machine-time! Not an equitable distribution of time at all. Shashi ________________________________________ From: [email protected] [[email protected]] on behalf of Bair, Raymond A. [[email protected]] Sent: Friday, March 27, 2015 11:27 AM To: LCRC Allocations Admins Subject: Re: [allocations-admins] Problem with the revised allocations spreadsheet I agree that we should do something to better to inform the committee. Many aspects can be improved,including distribution of time among organizations, visualizing time requests, etc. We also need to get rid of that too-complicated spreadsheet. Talking with Ti, I think we can build better tools once the userbase2 conversion is complete. Lets target the first FY16 allocation meeting to roll out some new capabilities. Ray ----------------------------------------- Ray Bair Computing, Environment, and Life Sciences Argonne National Laboratory and the University of Chicago TCS Building 240, Room 4122 9700 South Cass Avenue Argonne, IL 60439 email: rbair(at)anl.gov Phone: (630)252-5751 On 3/26/15, 7:53 PM, "Aithal, Shashikant M." <[email protected]> wrote:
I don't think the allocation committee should shy away from coming up with a policy to address this matter. Since every directorate is represented in the committee and the allocation process is open and transparent, there should be no reason why we should not try to implement some fairness doctrine. We could have an excel spreadsheet which shows the various directorates (and the divisions) and their total requested time for this half-yearly allocation - if there is disproportionate usage we let the committee see the disparity and judge for themselves. If some DD/ALD feels aggrieved, they can always make an appeal to the committee and justify why they should get preferential treatment. One could argue that cutting back on time for smaller projects could stifle burgeoning projects and prevent them from becoming the next "strategic" lab initiative.
Shashi ________________________________________ From: [email protected] [[email protected]] on behalf of Bair, Raymond A. [[email protected]] Sent: Thursday, March 26, 2015 5:39 PM To: LCRC Allocations Admins Subject: Re: [allocations-admins] Problem with the revised allocations spreadsheet
We judge each project by its individual merit. While there is a certain appeal to trying to balance utilization among groups, it would foment squabbles. Coming up with group or division or directorate shares depends on coming up with a fair metric. All the metrics (I can think of) for computing the shares are controversial. Some might say that a directorate's/division's/group's share should be proportional to the overhead they bring to the Lab, since LCRC is funded by overhead. Others might want a measure of how much computing their researchers are funded to do, e.g., programmatic and LDRD funds. Then people would argue that some have other computing resources and should lose share. It gets messy fast.
On the other hand, pushing some projects onto Fusion is a plausible idea. It would be good to propose two scenarios (1) without pushing projects to Fusion, and (2) with pushing projects to Fusion. If we provide SprayFlameLES with 500K instead of the requested 2.85M, then we have 8.7M hours more to allocate on Fusion, enough to make up the difference. If we do this we need to tell certain projects to run only on Fusion, and we will need a list of suggested projects before the meeting starts. Who could easily move? Most LCRC staff projects would need to be on the list.
Ray
On 3/26/15, 4:21 PM, "Aithal, Shashikant M." <[email protected]> wrote:
Ray:
I think, we should take a closer look at groups who use large allocations and scale them back more severely, or we don't leave Fusion exclusively for industry projects but open it up to everyone.
Shashi ________________________________________ From: [email protected] [[email protected]] on behalf of Bair, Raymond A. [[email protected]] Sent: Thursday, March 26, 2015 4:06 PM To: LCRC Allocations Admins Subject: [allocations-admins] Problem with the revised allocations spreadsheet
All of the formulas for sums (row 162) are incorrect. The correct total for column AE (New Decisions) was 34,597,000 not 20,797,000!
The value in cell A3 is incorrect. Should be 2015. (Makes no difference since 2014 and 2015 are not leap years.)
Most of the industry projects are listed twice, on both the Data page and the Industry page.
biodiesel-combustion
CAT-engine-modeling
cavitation-modeling
dual-fuel-sprays
LES-Engine
LES-sprays
We should also move the new project spray-flame-les to Industry (2.85M request). Correct?
What should we do now? I have suggestions bsed on past practice. See attached spreadsheet for numbers.
Reducing ApsRenewalLattice to 2M, and OPV to 500K.
Doing the above changes the total to 27,497,000. Now 26% over-allocated.
If we keep the priority 0-2 projects as they are, and scale the rest down to no less than 100K (for the 2nd half), then the scale factor is 71%.
See attached spreadsheet.
My revised total, percent allocation (relative to 100%), and scale factor are at the bottom of column AE
Comments and suggestions?
Ray
----------------------------------------- Ray Bair Computing, Environment, and Life Sciences Argonne National Laboratory and the University of Chicago TCS Building 240, Room 4122 9700 South Cass Avenue Argonne, IL 60439 email: rbair(at)anl.gov Phone: (630)252-5751
_______________________________________________ allocations-admins mailing list [email protected] https://lists.lcrc.anl.gov/mailman/listinfo/allocations-admins
_______________________________________________ allocations-admins mailing list [email protected] https://lists.lcrc.anl.gov/mailman/listinfo/allocations-admins _______________________________________________ allocations-admins mailing list [email protected] https://lists.lcrc.anl.gov/mailman/listinfo/allocations-admins
_______________________________________________ allocations-admins mailing list [email protected] https://lists.lcrc.anl.gov/mailman/listinfo/allocations-admins