AnnArborNotesMay2008

Background

• Meeting agenda: Tier2/Tier3 Workshop at UM (Ann Arbor)

Discussion

• User data management, "backing up" user-data. What is the Facility-strategy for dealing with this issue? We discussed approaches for storing user's precious data. BU is backing data to a tape system. Could there be a global system?
• Major infrastructure hole is lack of storage.

Networking

• iperf-sonar deployment on US ATLAS sites - goal. Can this be included in the NDT disk?
• FDT - can this be used rather than gridftp? There has been FDT-dCache work.

Operations

• LFC migration needs to be decided quickly.
• Worried about lack of users doing analysis, to stress the systems? Why? Data availability?
• A set of data management utilities needs to be created.

Space tokens in xrootd

• implemented in xrootd as virtual partions

High availability

• Shawn is evaluation a system based on iSCSI, VMs and a set of matched services to existing critical ones (globus, condor, dcache, afs, nfs)

Intel/AMD benchmarks

• Harpertown a big improvement over Clovertown and faster than latest Barcelona for simulation and reconstruction. No significant difference server vs blade.

E2E monitoring

• perfSONAR
• A collaboration of network operators; an architecture and set of protocols, and a deployed measurement infrastructure.
• Web-services based, locally controlled. Being standardized in the OGF.
• We would be the first community (non-network backbone) to deploy
• Updates to SW every 6 mos.

Intel roadmap

• Tick-Tock --- next is Nehalem

Dynamic Circuits

• A hybrid network where an IP network is not the best thing for the application. A bandwidth-hungry app over a limited time frame.
• IP bypass
• TeraPaths or LambdaStation as intermediaries.

Benchmarks of Dell disk

• See configuration and measurements from Wenjing
• System - Dell 2950 plus 4 MD1000 shelves, Raid60 in two partitions
• Varying multiple parameters, eg. number threads, readahead in the kernel, etc.

Physics perspectives

• There are support questions for Tier3 sites - what are the expectations for the US ATLAS Tier1 and Tier2?
• Use CDF/Tevatron as a reference for analysis - how is the model different in ATLAS, etc.
• Heavy reliance on the CDF CAF, then heavily interactive at Tier3 locally.

Tier3 and Analysis

• See Amir's talk

PROOF

• Ofer's presentation gives a simple tutorial for setting up a PROOF system
• Neng - focus on condor scheduling over shared resources and data management

wlcg-client and atlas-client

• Marco has version of osg-client extended with LCG tools and minimal LFC compat libs for use with gLite sites.
• ATLAS client includes fixes for new dq2-client package

Milestones

• FDR2
• 200 MB/s T1-T2
• OSG 1.0
• LFC eval complete
• WLCG - SAM/RSV, reliability availability metric (>80%)
• Provisioning of capacities according to pledges until 09/15
• Network performance and monitoring (i.e. get the PerfSonar? monitoring infrastructure installed and integrated into facility operations - 06/30)
• Define and set up a "sample" Tier-3 center - whitepaper
• Analysis benchmarks 100/200/500/1000 simultaneous jobs

Next meeting

• August 20-21 @ BNL: focus on the Tier3 center

-- RobertGardner - 27 May 2008