December, 2009 David Hart. Allocation Stats Processing Interfaces.

24
December, 2009 David Hart

Transcript of December, 2009 David Hart. Allocation Stats Processing Interfaces.

Page 1: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

December, 2009David Hart

Page 2: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Allocation Stats Processing Interfaces

Page 3: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Request and Allocation Trends

Total Avail

Page 4: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Dec. 2009 TRAC Meeting (“MRAC Cycle”)

Page 5: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Startups:

Page 6: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Plenary Session for “top 10” requests. Parallel Sessions for Others

Two Sessions, A Chair for each session Minimal Overlap (no need to attend both

sessions) PHY/AST/ATM/CFD/ASC CHE/MCB/DMS/DMR

Awards entered into common spreadsheet Google Doc Private document, accessible only by invitation.

Considerable Time Savings

Page 7: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Parallel Sessions Sept. 2009 TRAC Meeting

PHY/AST/ATM/CFD/ASC FOS Session 35 Requests; 105M SUs Requested

CHE/MCB/DMS/DMR FOS Session 39 Requests; 105M SUs requested

Dec. 2009 TRAC Meeting PHY/AST/ATM/CFD/ASC FOS Session

33 Requests; 102M SUs Requested CHE/MCB/DMS/DMR FOS Session

38 Requests; 103M SUs requested

Page 8: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

* RequiredForms

POPS development team is always improving, and maintaining interface.

Page 9: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Recent improvements Auto-fill

Supporting Grants and Co-PI Information is now automatically “refilled” on renewal requests (supplements and justifications, too).

Confirmation of auto-fill now requiredUpdate PI InformationAdd/remove CoPIsAdd/remove new/expire grantsModify Supporting percentage

Page 10: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Document Upload improvements (in progress) Single upload interface for all required docs (Simple) Selection of Document type

Main Document, Progress Report, CV, co-PI CV, TG-related Publications, References,

Uses Submit button below entry form (no more confusion with “Save to Date”)

Upload date now appears in document list(no more confusion about revisions)

Page 11: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

System Selection (in redesign phase) Present entry form is cumbersome

(must scroll through pages of entry form or use index at top)

Re-evaluate necessity of collected data Redesign input fields

(& include comma notation in numbers)

Page 12: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Richard Moore

Page 13: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

RAAR Report– (Recommended and Available Allocations Reconciliation)

Recommended Procedures for Handling Oversubscription

General Background Recommendations

Solving the legacy MRAC/LRAC Cycle Problems Review Process / Reconciliation Defined Process to Adjust Recommended Allocations for

Oversubscription

Page 14: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

REQUESTS = Original requests submitted from users RECOMMENDED = Awards recommended by TRAC

committee members based on merit review AVAILABLE = Amount of time made available by the

RPs for that allocation cycle ALLOCATED = Awards actually made, based on both

TRAC recommendations and availability (‘reconciliation’ or ‘normalization’ process) Includes overall availability as well as specific machines

Page 15: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Allocation award totals have traditionally been 60-70% of the Request totals.

Sept. TRAC Meeting: Requests = 810M, Available = 300M. Ouch!

Large differences in Recommended and Available Allocations require a mechanism to reduce Recommended Allocations to accommodate the available SUs (remove potential oversubscription).

Page 16: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Legacy Large and “Medium” request cycles persist (MRAC/LRAC quarterly and ½-yr cycles )

There is no simple way to normalize reductions across quarters (available and recommended allocations have to be reconciled at each meeting).

Request totals are difficultto predict. (From Oct. 2008 to July 2009 oscillations seemed to

dampen.)

Page 17: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Eliminate MRAC/LRAC waves 90% of the requests request are “MRAC” size 10% of the requests are “LRAC” size Recommend: Pro-actively re-distribute LRAC-

type requests By extension (1 quarter) or by early renewal (1

quarter) TG staff contact specific users and ask them to

switch cycles

Page 18: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Keep Merit Review Process separate from Oversubscription adjustments Let reviewers do what reviewers do best–> Provide

Recommend Allocations based on merit: Appropriateness of Methods Efficient use of systems Appropriateness of Computational Research Plan Usage of previous allocations, publications

Allocations Officers take care of applying adjustments for oversubscription– a TeraGrid Problem

Page 19: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Reviewers are not apprised of Oversubscription factor.(This will provide more consistent reviews of merit.)

Reviewers can use Funding to determine PI ability to manage and apply appropriate support to accomplish work in the Computational Research Plan. Reviewer should be blind to funding agency. (This encourages PIs to report ALL funding.)

For non-funded requests, science is reviewed by TRAC (no change from previous)

Benefits Recommended Allocations – i.e. merit-reviewed demand - can be

reported to NSF and the community. Reconciling availability limitations is removed from merit review

process – no double jeopardy.

Page 20: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Adjustments will be applied across all requests, by a uniform process (as formulaic as possible). Availability on individual machines/classes is the complicating

factor NSF has decided to no longer single out PIs with NIH

funding for special restrictions on usage However, funding source (NSF v non-NSF) will be considered

Factors to be used in reducing recommended allocations to fit availability Funding source (preference given to NSF-funded research) Size of award (preference given to small awards) Across-the-board reductions

The details of how these factors will be applied are still being developed – and will be confirmed with NSF

Page 21: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.
Page 22: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Recommended Allocations can be reported to NSF Documents merit-reviewed demand

Oversubscription Adjustment (Reconciliation) criteria is removed from review process – no double jeopardy.

Funding support can be easily applied at Reconciliation stage.

Page 23: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Funding All non-NSF funded requests have equal

consideration(NIH limits no longer apply– a “fair field” for all.)

When Adjustments are applied for oversubscription, NSF has priority or preference: Adjustments for non-NSF funded projects (or

proportion of non-NSF funding) will have a larger reduction factor.

Page 24: December, 2009 David Hart.  Allocation Stats  Processing  Interfaces.

Formula:

R * (Fnsf + Fnon-nsf *R+ ) * Recommended Allocation

R = “global” Reduction factor

Fnsf = Fraction funded by NSF grants

Fnonnsf = Fraction funded by non-NSF grantsR+ = Additional Reduction for non-NSF