AstroGrid-D

1
AstroGrid-D
2
GAC-Grid / Astro-GRID-DWP 5 Resource Management
for Grid Jobs

• Report by Rainer Spurzem
• (ZAH-ARI)
• spurzem_at_ari.uni-heidelberg.de

3
WP5 Resource Management for Grid Jobs Tasks

• Task V-1 Specification of Requirements and
  Architecture (Months 1-9, AEI, AIP, ARI, MPA,
  MPE, ZIB)
• Start Sep. 05, Deliverable D5.1 May 2006
  OVERDRAWN!
• Task V-2 Development of Grid-Job Management
• (Months 10-18, ARI, MPA, ZIB)
• Start June 06, Deliverable D5.2 Feb. 2007
  COMING
• Task V-4 Adaptation of User- and Programmer
  Interfaces (Months 16-21, AEI, AIP, ARI, MPA,
  MPE)
• Start Dec. 06 Deliverable D5.4 May 2007
• Task V-3 Development of Link to Robotic
  Telescopes, Requests (Months 13-18, AIP, ZIB)
• Start Sep. 06 Deliverable D5.3 Feb. 2007

4
WP5 Resource Management for Grid Jobs

• Group Activities
• Members of gacg-wg5_at_aip.de
• R. Spurzem (Coord., ZAH), D. Elstner (AIP),
  O. Wehrens (AEI), H.M. Adorf (MPA), W. Voges
  (MPE), A. Reiser (TUM), T. Röblitz (ZIB), A.
  Beck-Ratzka (AEI), M. Hoegqvist (ZIB), T.
  Bruesemeister (ARI-ZAH)
• Deliverable 5.1 stalled after Telecon in June
  questions to be discussed
• job language
• relation to other middleware
• provision of resources / brokerage

5
Job Languages

• JSDL Job Submission Definition Language
• Global Grid Forum, XML based,
  interoperability
• between schedulers, contains
• Resource Requirements
• Dependencies to other Jobs
• Relation to datasets
• Architecture of job (?)
• JSDL Doc. handed to scheduler, similar, RSL,
  x-RSL
• (check whether translator exists)

6
Job Scheduling
7
Job Submission Logical Structure
Job Description

• Software Description Use Case, Code, Compiler
• Hardware Description Requested Hardware
  (Special?), Parallelism
• Job Control Flow Description Batch Script,
  Unicore-Control-Flow, GAT, JSDL, (x)-RSL?
• Data Access Description Linked to Job
  Control, to Data, Metadata

Job Submission

• Resource Broker User Supplied Data Walltime,
  Memory, Checkpoints,
• Informations
  from Job Description
• Informations
  about available resources

Job Control

• Status Informations User/Site Execution Status,
  Time, Queue Status
• User Notification Job Termination, Checkpoint,
  save job metadata?
• User Interaction with Job Link WG6

8
Example for Job Control Flow of NBODY6 Case
9
Open Questions
How to implement complex workflows, such as ProC
? JSDL/RSL good for atomic jobs Need GACG
specific job environment (cf. DEISA) Other
Alternatives Unicore but less flexible? Need
to point to firewall problems Broker links to
RDL Resource Description Language
10
What is there I?

• hydra non-GRAPE cluster (20 P4 2.4Ghz)
• (SFB439Land BW)
• titan GRAPE cluster (64 P4Xeon 3.2 Ghz, 32 G6 125
  Gflops each, 32 Xilinx FPGA Virtex-4 (not yet
  there), total 128 CPUs, 4 Tflops peak)
• (Volkswagenstiftung, SFB439, Land BW)

11
How to proceed?

• D5.1 Architecture Specifications
• some specs. may need use case
• input
• Proceed with implementation of pioneer use cases
  (to D5.2)