Скачать презентацию CMS Computing Xiaomei Zhang CMS IHEP Group Meeting Скачать презентацию CMS Computing Xiaomei Zhang CMS IHEP Group Meeting

7d8d2bac9438fa1870e2d050d2e6ad5a.ppt

  • Количество слайдов: 13

CMS Computing Xiaomei Zhang CMS IHEP Group Meeting December 14 2008 CMS Computing Xiaomei Zhang CMS IHEP Group Meeting December 14 2008

Outline • • • T 3 resource and discussion T 2 resource and discussion Outline • • • T 3 resource and discussion T 2 resource and discussion CRAB and Local Batch Submission functions CRAB Server Site monitoring 2

T 3 resource and discussion • CPU resource (96 cores) – Old: 6 machines, T 3 resource and discussion • CPU resource (96 cores) – Old: 6 machines, 48 cores; New: 6 machines, 48 cores • Storage resource – cmsd 01 and cmsd 02 (old: /cms 1, /cms 2, /cmsmc 1, /cmsmc 2) – /home/cms/xxx(old: /home/lhc/cms/xxx) – Backup: /home/cms/xxx, /cmsd 01/public/cms-software(<100 G) • Login – cmsui 01, lxslc(old lxplus, redhat) • Plan – Balance between local disk and SE • Directly read and write SE after SE increase to 100 TB – Shut down VNC to use resource more reasonably and efficiently 3

T 2 resource and discussion • CPU resource (468 cores) – Old: 104 cores; T 2 resource and discussion • CPU resource (468 cores) – Old: 104 cores; New: 364 cores – Shared with Atlas, biomed, cms • CMS 50% production, 50% analysis – In theory, 100 or so for analysis(including physics group) – In practice, 200 or more , you can get “running and queue” information from • http: //ganglia. ihep. ac. cn/ganglia/addons/job_monarch/? c=BEIJING-LCG 2%20 WN • Storage resource (20 TB ->100 TB) • Plans – Give local users higher priority • Establish special role for local users in CERN voms (/cms/Role=NULL->/cms/Role=Beijing) • 100 or less – SE will be too high in the future • Phedex transfer(debug+prod), production, global analysis, local analysis 4

T 2 resource and physics group • A nominal T 2 resource: – ~500 T 2 resource and physics group • A nominal T 2 resource: – ~500 batch slots, 200 TB – ~ 468 batch slots(sharing) , 100 TB – 200 batch slots in T 2_CN_Beijing • CPU – Half for simulation and half for analysis – 100 for analysis in T 2_CN_Beijing • Storage – – – Central Space 30 TB(? ) MC space 20 TB Physics Group Space 60~90 TB Local Storage Space 30~60 TB User Space 1 TB (20 members) 100 TB in T 2_CN_Beijing soon( MC space 20 TB, Local Storage 30~60 TB, User 5 Space 20 TB, tmp 1 TB)

T 2 association to physics group 6 T 2 association to physics group 6

Should we join physics group? • T 2 physics group meeting this workshop • Should we join physics group? • T 2 physics group meeting this workshop • Advantage: – Share cpu and space resource • Higher priority – More support • Crab server • Stage out to an official area – More experience • Space management • Batch slots management – From physics point of view, … • Disadvantage: – Requirements on resources 7

CRAB and Local Batch Submission functions • CRAB used for local batch submission – CRAB and Local Batch Submission functions • CRAB used for local batch submission – Plan to release next year – Still a plan, no real actions until now • The functions included in the future – Support the input dataset reading from and the output dataset writing to the local disk – Support dataset registering in local DBS – Read or Write to SE with certificate or without certificate? • Still in discussion 8

CRAB Server • standalone and client-server mode • client-server mode Grid – a server CRAB Server • standalone and client-server mode • client-server mode Grid – a server is placed between the user and the Grid – Encourage analysis users to use it • Standalone mode – Directly submit to grid as now – Maintain or not support in the future? Crab server • Still in discussion – Used for local batch submission? • Still in discussion • guide and questions: – https: //twiki. cern. ch/twiki/bin/view/CMS/SWGuide. Crab – Get support from hn-cms-crab. Feedback mailing-list 9

CRAB Server • Goals: – make users to submit a large number of jobs CRAB Server • Goals: – make users to submit a large number of jobs easier • • don't wait for submission, strong resubmission functionality email you when jobs are finished know the job status easily in the web page No limitations on the size of inputbox – Easy to maintain, extend, support • crab client would be kept as stable as possible, as simple as possible • crab server responsible for most of maintaining and extending tasks(physics group) • Status: – The first release in May this year • 14 May 2008: CRABSERVER_1_0_0 has been released – Two crab servers available now • Bari and Pisa • Servers will be managed by physics groups • A common server in CERN set up soon 10

The progress in batch submission • advantage and disadvantage – Easy for functions control The progress in batch submission • advantage and disadvantage – Easy for functions control and debug – Difficult to maintain in the future, time consuming • plans and problems – I plan two support ways – Support the dataset reading from and writing to local disk • but can not support disk data to be published in DBS – Support the dataset reading from and writing to SE • require dcap open to nodes without certificate • or require srmcp with certificate 11

Site Monitoring • New transferring link tests – At the rate of 2. 5 Site Monitoring • New transferring link tests – At the rate of 2. 5 Mb/s – Too many parallel links keep SE busy • Information counted in site monitoring – CMS SAM availability – CMS Job. Robot – CMS Commission links • Methods used in site monitoring – Site flag: • UNCOMMISSIONED: site unusable or very unstable • WARNING: unusable today but ok before • COMMISSIONED: usable and stable – The flag for the day depends on the metrics of the previous 7 days 12

Site Monitoring 13 Site Monitoring 13