Discussion - Analysis of existing products and task alocation - Wednesday morning ========== * VO - overseen by Ian Fisk - authorization, sharing of others' data [Marcus Lorch, Va Tech] /- make application VO-aware, extend VOMs [M. Lorch, C. Steenberg, Tech-X | (auth w/ CAS)] |- Authorization <-> Job Execution Service [M. Lorch, C. Steenberg, Tech-X | (auth w/ CAS)] |- Inter-VO management (eg: EDG/US-CMS) [M. Lorch, C. Steenberg, Tech-X | (auth w/ CAS)] | - VOs must be able to accept each others' VOs, even if managed | differently. |[CHEP03 paper by C. Tull et al.] |- Vox project: US-CMS, US Atlas, BNL, FNAL, Indiana | \-> [M. Lorch, C. Steenberg, Tech-X (auth w/CAS), C. Tull, T. Levchina, R. Baker, R. Gardner, Alos Frohner] * Software Install - A. Roy (w/ Boock Joo, N. Ratnicova, Saul Youssef, Flavia Donno) - Config Management - VDT/GDB/PACMAN * Virtual Data System - Rick Cavanaugh (w/ Mike Wilde, Eva Deelman, Greg Graham (runjob, interface to S?M), Jim Branson [methods]) - Griphyn, PPDG, LVDGL - Interface to global and local planners <=> Job Execution Service [SPHINX (UFL), Sanjay Ranka] * Data Service Catalog (Metadata) - Joe Perl (CS11) - Web/Grid service that can have different implementations to talk to BaBar, etc. for providing something that applications can read. - Query to get handle to pass to planner - Application interface to VDS * Replican Management Service + Data Mover + Delivery Service - Joe Perl (CS11) - Replica Location Service => {Physical replicas} - data managers pick the best source from the list returned by the RLS /- After RMS, then authorization, policy, optimization \- Timeouts while building list \ | -> Estimators |-> Replica Selection Service (New!) - Replica Selection Service still needs a replica deliver service architecture RSDS <-> RGMA; MDS; MonaLisa Some disucssion on the actual role of the planner. Should the planner simply schedule jobs, reserve resources, submit jobs, or some combination of these? It's difficult to have an automated Replica Selection Service and Job Planner that doesn't take into account CPU availability as well as data location. * Concrete Job Planner - Interfaces to RMS, Site Planner, RSS, and Execution Location Service. RSS and ELS are closely tied. * Job/Process Execution Service (JP == Job/Process) Iosif Legrand, H. Pi - Don't have an architecture yet /- JP status and control |- Lifecycle according to queueing system or operating system |- Control communication |- JP / client interaction | - supports queries, responses, load python code | |-> JP Capability - Realtime monitoring; performance tracking - {Failed JP} ---------\ \__ Restrategize => Re-execute => Workflow / Monitor/Track | Steering and | Optimization - {Performace too low}-/ [Redirection] * Data Collection Services (new!) (by subjob) - Physics Data Product -> Logical Data Product -> {Expression} - This is part of the application - Some "system" collection services * Estimators - External and Internal * Events * Adaptive Learning