CSF4 MetaScheduler

1
CSF4 Meta-Scheduler

• Zhaohui Ding
• Zhaohui.ding_at_email.jlu.edu.cn
• College of Computer Science Technology
• Jilin University

2
Agenda

• What is Meta-Scheduler
• What is CSF
• CSF4 Architecture
• Integrate CSF4 and Gfarm
• Demo Practice
• Future Work

3
What is Meta-Scheduler
CSF4
GRAM
4
Meta-Scheduler VS Local Scheduler

• Local Scheduler
• For Cluster Environment
• Same Hardware Architecture, OS
• Typically different Resource Management softwares
  (RMs) like LSF, PBS, and Sun Grid Engine are
  responsible for load balancing and resource
  sharing within each local administrative domain.
• Meta-Scheduler
• For Grid Environment
• Different Hardware, OS, Infrastructure, Local
  Scheduler, Resource Manager
• In a Grid-computing environment, there is a
  common requirement for users to query, negotiate
  access and manage resources existing within
  different administrative domains at Grid level.
  The meta-scheduler is designed to perform such
  global wide policies.

5
Meta-Scheduler and Local Scheduler in Grid
6
CSF4 Summary

• What is CSF
• Community Scheduler Framework
• CSF4 is a group of grid services host in GT4
• CSF4 is the first GT4-WSRF compliant
  Meta-Scheduler
• Open Source project and can be accessed at
  http//sourceforge.net/projects/gcsf
• The development team of CSF4 is from Jilin
  University, PRC
• CSF4 has been added to Globus Toolkit 4 as an
  Execution Component

7
CSF4 in Globus Toolkit 4
8
CSF4 Architecture
9
CSF4 Functionalities

• Submit, control and monitor jobs at a grid level.
  
• Queue Service for jobs is provided
• Choose satisfied cluster for jobs automatically
• Create and manage advanced reservations for LSF
  cluster.

10
CSF4 FunctionalitiesResource Supported

• Infrastructure Supported
• WS GRAM (GT4 GRAM)
• Pre-WS GRAM (GT2 GRAM)
• Local Scheduler Supported
• LSF (support advanced reservation)
• PBS
• SGE
• Condor

11
CSF4 Functionalities Scheduling Policy
Framework

• A scheduling policy framework is provided so that
  site and user specific scheduling policies can be
  implemented regardless of the underlying resource
  manager.
• User can configure scheduling policies for queue,
  and even write a new scheduling policy as a
  plug-in without modify the source codes of CSF4.

12
Existent Scheduling Policies Plug-in

• FCFS plugin
• First Come First Serve
• Default policy
• always loaded by CSF
• Throttle plugin
• An optional plugin
• Constraint the number of jobs lunched in one
  scheduling cycle

13
Schedule plugin scheduling policies

• Scheduling polices are defined in a queue
• Each policy is implemented inside a scheduling
  plugin module
• A queue can load multiple plugin modules

14
CSF4 Functionalities (cont.)Deploy CSF4 in a
Grid Community

• Multiple CSF4s Can Exchange Cluster Information
  Each Other

15
Deploy CSF4 in a Grid Community

• Example

16
Integration of CSF4 Gfarm
A Scalable Grid Infrastructure For Data Intensive
Applications
17
Integration of CSF4 Gfarm
gfsd
host1
host3
host2
host5
host4

• Regardless of where the data is locate

18
How to get CSF4?

• Release Page (Recommended)
• http//sourceforge.net/project/showfiles.php?group
  _id103105
• CVS
• http//sourceforge.net/cvs/?group_id103105
• Globus Toolkit 4 Package
• In contrib directory

19
Installation

• Requried Software
• Globus Tookit 4
• gpt-build csf-4.0.1-src.tar.gz
• gpt-postinstall

20
Configure Meta-Scheduler

• Edit GLOBUS_LOCATION/etc/metascheduler/metaschedu
  ler-config.xml
• Example

21
Configure Resource

• Edit GLOBUS_LOCATION/etc/metascheduler/resourcema
  nager-config.xml
• Example
• ltclustergt
• ltnamegt sge-32 lt/namegt
• lttypegt GRAM lt/typegt
• lthostgt rocks-32.sdsc.edu/jobmanager-sge lt/hostgt
• ltportgt 2119 lt/portgt
• ltversiongt2.4lt/versiongt
• lt/clustergt

22
Services started by GT4 container (cont.)

• globus_at_rocks-32 gt4 globus-start-container
• 1 https//198.202.88.328443/wsrf/services/Trig
  gerFactoryService
• 2 https//198.202.88.328443/wsrf/services/Dele
  gationTestService
• 9 https//198.202.88.328443/wsrf/services/met
  ascheduler/QueuingService
• .
• 16 https//198.202.88.328443/wsrf/services/met
  ascheduler/ResourceManagerFactoryService
• 23 https//198.202.88.328443/wsrf/services/met
  ascheduler/ReservationService
• 25 https//198.202.88.328443/wsrf/services/Ind
  exFactoryService
• 26 https//198.202.88.328443/wsrf/services/Rel
  iableFileTransferService
• 47 https//198.202.88.328443/wsrf/services/met
  ascheduler/ResourceManagerLsfService
• 49 https//198.202.88.328443/wsrf/services/met
  ascheduler/JobService
• 56 https//198.202.88.328443/wsrf/services/Man
  agedJobFactoryService

23
Demo Practice

• Before testing
• grid-proxy-init
• cp zding/practice/ .
• scp zding/practice/ USER_at_rocks-170

24
Demo 1 Query Available Cluster

• csf-job-RmInfo

25
Demo 2 Submit WS GRAM job

• csf-job-create rsl job.xml -name zding_job1
• csf-job-submit zding_job1
• content of job.xml
• ltjobgt
• ltexecutablegt/bin/echolt/executablegt
• ltdirectorygtGLOBUS_USER_HOMElt/directorygt
• ltargumentgtthis is a globusrun-ws
  samplelt/argumentgt
• ltstdoutgtstdoutlt/stdoutgt
• lt/jobgt
• csf-job-data zding_job1
• cat /stdout

26
Demo 3Submit pre-ws gram job

• csf-job-create rsl job.rsl name zding_job2
• csf-job-submit zding_job2
• content of job.rsl
• (executable/bin/echo)
• (directory(HOME))
• (arguments"this is a globurun example")
• (stdoutstdout2)
• csf-job-data zding_job2
• cat stdout2

27
Demo 4Submit job with CSF4 Gfarm

• Application MEME353
• csf-job-create rsl job3.rsl name zding_job3
• csf-job-submit zding_job3
• csf-job-data zding_job3
• gfarmfs lsfu /gfs/USER -o kernel_cache
• ls ltr /gfs/USER/USER

28
Demo 5Submit multiple jobs to CSF4

• for i in 1 2 3 4 do echo i csf-job-create
  -rsl job4.rsl -name jobi done
• for i in 1 2 3 4 do echo i csf-job-submit
  jobi done
• for i in 1 2 3 4 do echo i csf-job-data
  jobi done
• qstat

29
Future Work

• CSF4 Portlet
• Base on GridSphere
• CSF4 Opal Integration
• Opal Wrapping Scientific Applications as Web
  Services
• Implement data-aware scheduling policy in CSF4
• Scheduling policy for data intensive application
• Improve the performance
• Workflow supported

30
Thank you