1 / 10

UB Overview

UB Overview. Role of the UB Represent user community within GridPP management Request and allocate Tier-1/2 hardware resources Quarterly meetings (Q2 meeting this afternoon) Meetings are open to all collaborators Current issues Resource allocation: Tier-1 resource contention

xarles
Download Presentation

UB Overview

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. UB Overview • Role of the UB • Represent user community within GridPP management • Request and allocate Tier-1/2 hardware resources • Quarterly meetings (Q2 meeting this afternoon) • Meetings are open to all collaborators • Current issues • Resource allocation: Tier-1 resource contention • GridPP3 requirements definition • Migration to CASTOR2 at Tier-1 (discussion this afternoon) Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  2. Resources: T1 CPU • Available in July: 1031kSI2k • Farm is essentially fully used • Note that efficiency is still an issue - investigations continue Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  3. Resources: T1 Disk • Available in July: 179TB • We are critically short of Tier-1 disk • Deployment of new resources delayed by infamous firmware issues Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  4. Resources: T1 Tape • Available in July: 532TB • No further expansion of ‘old’ tape technology planned Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  5. Resources: T2 • T2 accounting still needs to be optimised… • Available in July: 6322kSI2k; 887TB (!) • Note take up ratio… Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  6. GridPP3: Resources • Process • All experiments asked for forward look to 2012 • No explicit constraints placed on requirements • Future activities (not yet active in GridPP2) also gave input • LHC Experiments • Tier-1 resource balance informed by computing models • Tried to distinguish classes of use for CPU (not yet for disk) • Tier-1 scale set by community size or by practical minimum (CMS) • Tier-2 scales as community size • Resources increased significantly w.r.t current MoU numbers • Tier-1/2 resources roughly 1:1 (empirical confirmation of Doyles nth Law) • Other Experiments • Envelope of 5% T2 disk/CPU + T1 tape should cover the requirement • Future linear collider work may need more beyond 2010 • UB will continue in allocating role in GridPP3 Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  7. GridPP3 Resources Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  8. GridPP3: UB Role • Issues with current UB structure • Balance between formal representation / open meetings • Narrow point of contact with the rest of GridPP • Quarterly resource allocation is increasingly impractical • A more hands-on approach is required for T1 allocation • Optimisation of Tier-2 resource allocation process • Load on UB chair • Proposals for GridPP3 • Role of UB chair should be reviewed • Empowered to make practical short-term decisions on resources • Cross-membership with new Deployment Board • More proactive discussion of technical requirements in the UK • As opposed to top-down approach via WLCG Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  9. GridPP3: Data Flows • Outstanding item for GridPP3 planning: • Document our understanding of inter- and intra-site dataflows • Informs hardware, infrastructure and network planning at T1/T2 • Most information exists, needs to be collated and scaled for UK centres • Experiment dataflows: • T0 -> T1; T1 <-> T1 (well understood?) • T1 -> T2 (potentially very large, not completely understood) • E.g. what is the requirement for T1 <-> T2_ext and vice versa? • Disk <-> MSS bandwidth (becoming understood) • Simplifying assumption is to ban ‘random’ tape access • Disk <-> CPU bandwidth • Understood for bulk data operations at T1 • Large uncertainties for analysis use - affects both T1 and T2 • WLCG working group (incl. NB, RJ, DN) will examine T1/T2 questions Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

  10. Other Issues / Summary • Upcoming issues: • CASTOR2 testing / migration planning at Tier-1 • UK resource planning for dataflows (T1 and T2) • CPU efficiencies, including new architectures • E.g. benchmarking on multicore - what scaling will we see? • Summary • UK resources are being heavily used • Contention for resources (esp. T1 disk) will soon hit us hard • GridPP3 requirements planning well advanced • Work to do on understanding dataflows (short and long term) Dave.Newbold@cern.ch 16th GridPP Meeting, 28/6/06

More Related