RAL Tier1 Resources Review 20101006

From GridPP Wiki
Jump to: navigation, search

Documents

The following documents were circulated before the meeting:

Attendees

Andrew Sansum, David Corney, Matt Hodges, Martin Bly, Dave Britton, Sarah Pearce, John Gordon, Chris Brew, Alastair Dewhurst, Tim Folkes, Gareth Smith, Matthew Viljoen, Andrew Lahiff, Roger Jones

Notes

Overview of purpose of meeting, review agenda and meeting mechanics (RAS)

The Tier-1 Resources Review meetings have been proposed in order to close the loop between experiment resource requests, usage patterns, Tier-1 capacity planning, financial modeling/needs and MoU commitments. In each meeting current usage as well as disk, tape and cpu capacity will be reviewed, and the forward projection of the capacity plan and commitments will be considered.

  • Dave Britton commented that the more information about resource requirements that the experiments can provide the better.

Tier-1 deployed capacity and near term projections (includes phase outs and procurements in flight)

Matt Hodges presented a comparison of available resources with MoU commitments for the years 2010-2011 for cpu, disk and tape. WLCG pledeges are being met for CPU with additional capacity available for the non-LHC experiments. WLCG pledges for disk are currently met, and either the SL09 or 2010 disk procurements will meet the 2011 pledges. A shortfall in planned tape provision was noted for Q2 2011, however current levels and growth rate indicate that there will be sufficient tape capacity to meet experiment requirements.

Experiment resource usage

Andrew Lahiff presented plots showing usage of the LHC and other VOs over the past 12 months for CPU, disk and tape. ATLAS, CMS and LHCb hadn't ever reached their CPU allocation, while ALICE have exceeded their allocation in each of the past 3 months, including July where they exceeded their allocation by around 17 times. The significant increase in ATLAS disk usage from 0.6 PB (Jan) to 1.75 PB (Sep) was noted. Alastair said that part of this is due to ATLAS sending RAW data to all Tier-1s. Some time after heavy-ion running ATLAS will stop writing RAW to disk, freeing up about one third of the used disk. CMS disk usage has been relatively constant over the past 12 months. LHCb's disk usage over has tripled since April, and is now the VO which is closed to reaching its disk allocation. As expected CMS has been the largest user of tape, and over the past 12 months CMS tape usage has increased from less than 1 PB to around 1.7 PB. ATLAS tape usage has remained small, and they are still have over 1 PB of headroom. The Tier-1 will improve monitoring of tape compression of the different VOs. It will be the Tier-1’s responsibility to run repacking when necessary. If the compression ratio of a particular VO begins decreasing, this will be investigated before a crisis can occur.

Experiment requests and proposed allocations

No small VOs were present. It was noted that allocations for the non-LHC VOs haven't changed since December last year, but there have been no issues with disk or tape usage. Use of SL09 disk servers may be planned at the next Resources Review meeting (December). Best use should be made of the SL09 disk servers, but VOs must be aware that any allocations will be part of the 2011 pedges delivered early. It was agreed that experiment requirements for the next 3 months should be gathered in a consistent manner before each meeting, e.g. a questionnaire sent out by the UB. It’s important to get information from the small experiments in case their requirements change significantly.

Summary of MoU commitments - current status - do we meet pledge - planned to commit

See the discussion on capacity and near term projections.

Financial situation and longer term capacity planning

  • (DB) The big unknown is the Comprehensive Spending Review. This won’t have an immediate effect, but any impact will be further downstream after 2012.

Open Actions

Action ID Priority Owner Action Status
20101006-01 Medium Andrew Sansum Speak to Glenn and decide if there is any useful information he might be able to gather for us from the experiments for each 3 month planning window.