Testbed Status Charles Cal Loomis CNRS charles loomiscern

  • Slides: 6
Download presentation
Testbed Status Charles (Cal) Loomis (CNRS) charles. loomis@cern. ch Project Technical Board CERN October

Testbed Status Charles (Cal) Loomis (CNRS) charles. [email protected] ch Project Technical Board CERN October 2, 2002 C. Loomis – Testbed Status – Oct. 2, 2002 – 1

Application (Production) Testbed Deployment v. EDG release 1. 2. 2 deployed on six sites.

Application (Production) Testbed Deployment v. EDG release 1. 2. 2 deployed on six sites. —CC-IN 2 P 3, CERN, CNAF, Karlsruhe, NIKHEF, RAL. v. Around 15 others waiting to join. —Other grid projects (Cross. Grid, Grid. PP, INFN Grid, …) —More still once some national CAs are approved. v. Start to see real computing power, storage space on testbed. Limitations v. GASS cache problem Low-rate submissions only. v. Grid. FTP bug 20 minute time limit on file transfers. v. MDS Extremely unstable. Release 1. 2. 2 is NOT suitable for widespread deployment. C. Loomis – Testbed Status – Oct. 2, 2002 – 2

Recent Problems MDS Instabilities v. Top problem knock-on problems for users. —problems with match

Recent Problems MDS Instabilities v. Top problem knock-on problems for users. —problems with match making —finding replicas v. Moved back to manual configuration of II. —Script used to monitor status of sites. Logging & Bookkeeping v. Buckled under very large number of simultaneous requests. v“Abuse” of dg-job-status –all. v. Can improve with purging database. C. Loomis – Testbed Status – Oct. 2, 2002 – 3

Testing for 1. 2. x Job Manager (GASS cache): v. Rate problem gone when

Testing for 1. 2. x Job Manager (GASS cache): v. Rate problem gone when using fork job manager. v. Bug found with pbs backend scripts. —reported to Globus and fixed —need to repeat testing v. Haven’t confirmed or denied: —EDG GK works with new job manager. —Combination works with WP 1 job submission. MDS: v. Dynamically-linked MDS 2. 2 works; gross failure modes gone. v. Statically-linked MDS is not possible because of LDAP. v. Will need to deploy parallel Globus trees to have this work. C. Loomis – Testbed Status – Oct. 2, 2002 – 4

Testing for 1. 2. x (cont. ) Grid. FTP: v. Statically-linked server/client OK (in

Testing for 1. 2. x (cont. ) Grid. FTP: v. Statically-linked server/client OK (in 1. 2. 3). WP 2 Software: v. Replica Manager recompiled/relinked. v. RM tested with new Grid. FTP servers. v. GDMP is being tested now. Configuration: v. Details very different for job manager. v. Other minor details like updating dependencies. v. Should have updated configuration rpms shortly. C. Loomis – Testbed Status – Oct. 2, 2002 – 5

Summary Application Testbed: v. See familiar problems with software. v. Number of users/jobs increasing

Summary Application Testbed: v. See familiar problems with software. v. Number of users/jobs increasing starting to see scalability problems. Towards 1. 2. x: v. Testing progressing (only job manager not verified). v. Configuration being updated to allow mixed release. C. Loomis – Testbed Status – Oct. 2, 2002 – 6