You are here: Home User Information Facility Services Frontier Frontier Meetings for ATLAS Meeting Minutes Archive 2011 Minutes: 4/13/2011

Minutes: 4/13/2011

by John S. De Stefano Jr. last modified Apr 18, 2011 08:47 AM
Notes from the ATLAS Frontier meeting on April 13, 2011.

frontier-minutes-20110413.txt — Plain Text, 3 kB (3827 bytes)

File contents

Participants: Alastair Dewhurst, John DeStefano, Dave Dykstra, Elizabeth 
              Gallas, Fred Luehring 
Apologies:    Dario Barberis, David Front

*** Site status: ***

- No news

- Peaks of high usage on Squids, with high number of hits; no load worth 
- Hits on CMS Launchpads from RAL T2 and T3 sites during weekend network 
  * Sites should have their own Squids, or share with installed ATLAS Squids
- Questions from ATLAS S&C:
  * Reporting down-time: currently not possible; may be addressed via AGIS 
  * Sharing Squids between VOs or services: not currently supported by ATLAS 
    but to be addressed in upcoming AGIS requirement

*** Testing: ***

LHCb (Stefan, Marco):
- Have been using local testbed at CERN provided by Dave
- Very interested in further testing on broader scale, starting with T1s and 
  possibly T2s later
- Would use deployment model more like CMS's than ATLAS's: central Frontier at 
  CERN and Squids at all other sites
- Anticipate traffic of about 100 MB per hour
- Have asked to use ATLAS Frontier infrastructure for testing
- Comments that production infrastructure could be shared quite easily between 

Geometry (Vahko):
- Tests went well; decision must be made internally whether to use Frontier

*** Deployment: ***

- Frontier/Squid can be monitored by shifters once instructions are available, 
  but tools need to be fixed/established first (Jarka)

IN2P3-CC (Ghita):
- Has a single Squid and does not see need to add more, as local users are 
  connecting directly to a very robust Oracle instance, and sees no reason to 
  alter this access pattern
  * Remote or T2/3 sites are using Frontier and Squid for access
  * Would pose a problem were database release production ever to be stopped
- At latest RPM version, which is behind the latest Frontier, Squid, Tomcat 
  open-source packages by multiple versions (same situation as CERN)
- Sees need for ATLAS packages and service support to become more generic for 
  additional services and VOs (possibly for WLCG?), especially in the case of 
  Squid, and more a more systematic approach to the service in ATLAS in the 
  form of a "critical" service

ATLAS packages:
- David progressing with improvements to existing packages

AGIS class feedback:
- Latest proposal:
- An additional parameter is needed for 'SquidService' class/table/entity to 
  denote the service for which the Squid will serve as a cache (e.g., 
  Frontier, CVMFS, ?)
- It would be nice to have an example of how sites would appear in these 
  fields, perhaps based on existing site values in TiersOfAtlas and simply 
  duplicating current ToA functionality, before committing to effectiveness 
  and accuracy of latest class diagram
- May require an additional parameter for Frontier servlet name, which differs 
  between sites and is part of access URL
- Not clear whether load balancing denotation for FSConfiguration has been 
  added, or why both endpoint and service are supported (both not needed)
- Not clear how to determine name of backup site for each site, or how 
  'is_backup' would be returned in the class diagram
- Not clear what 'stype' is in [Squid|Frontier]Service, or 'type' in Service 
- Not clear what the class diagram represents as far as methods, table 
  structure, API needs, monitoring specifics, loading test data from ToA

Dave's time commitment:
- Needs to minimize time spent on ATLAS and Frontier, concentrate more on OSG
  * Needs plan for Frontier support and time needed from Dave for next year

*** A.O.B.: ***

- Presentation at ATLAS Software & Computing Week:
- Dave will be on vacation the week of 4/25
- Alastair will be away until after Easter has arranged representation for RAL
Document Actions
Filed under: , , ,