Ian Bird CERN Rob Gardner University of Chicago
Ian Bird, CERN Rob Gardner, University of Chicago GRID MIDDLEWARE & TOOLS SESSION SUMMARY
Introduction 82 abstracts submitted, 36 oral presentations (7 sessions), 44 posters, [2 withdrawn] Categories: cover a broad range Experiment experiences Data Management Workload Management Monitoring, Information, Accounting Security & Authorization Fabric & Deployment
Experiment experiences
Grid reliability – Pablo Saiz
Grid efficiency during CMS data challenges – Oliver Gutsche
D 0 – reprocessing on OSG Amber Boehnlein Common theme: making sites reliable requires debugging sites/systems one by one
Job agents – pilot jobs Monitoring Alien grid environment - Pablo Saiz
Data management
SRM v 2. 2 – Flavia Donno 18 month effort to agree, build, test, deploy new version
d. Cache – one of several MSS systems -Patrick Fuhrmann – overview of d. Cache developments -- Gerd Behrmann – distributed instance for NDGF
LCG Data management tools LFC, DPM, FTS – Markus Schulz
Examples of services that consider deployment & management issues
CORAL – distributed database access Dirk Duellmann
Workload management
Pilot jobs?
Pilot jobs – and variants: Such a good idea – everyone wants one …
Stuart Paterson – optimizations in DIRAC Marianne Bargiotti Integrity checking in DIRAC
Pilots can move intelligence into the job Paul Nilsson – Panda experience
g. Lite WMS developments Marco Cecchi
Igor Sfiligoi – comparison of WMS CHEP'07, Victoria 21
Monitoring, information, etc.
Experiment dashboards Julia Andreeva Monitoring from VO/user perspective
Grid. ICE – monitoring Guido Cuscela Permits different views of running jobs
James Casey Advances in monitoring of grid services
Stephen Burke – 6 years experience with GLUE schema Martin Flechl – details on integration of information systems
Security, authorization, etc
David Groep - gl. Exec Supporting pilot jobs
Fabric & Deployment
Greig Cowan Using DPM over the WAN
Addressing failover for core operations services – Alfredo Pagano Various strategies
Platform LSF – Robert Stober Integrating heterogeneous clusters
Observations Solutions exist for most needs now – Certainly not all perfect yet Experiment layer relatively deep Plethora of workload management systems Not so many for data management … Service management issues starting to be addressed by some services (DPM, LFC, FTS, Gridsite, Coral) But in general little thought on how site managers should manage services Interoperability / interoperation
Observations Workload management Everyone wants pilot (aka glidein) jobs (and everyone has written a system to submit them) Commonality – to reach a reliable service experiments need to systematically debug sites being used: D 0, CMS, dashboards, … Sophisticated systems to monitor, debug, recover Dirac, dashboards, grid service monitoring, etc. , To improve reliability and help debug the system
- Slides: 34