Heuristics for Scheduling Parameter Sweep Application in Grid environments - PowerPoint PPT Presentation

About This Presentation
Title:

Heuristics for Scheduling Parameter Sweep Application in Grid environments

Description:

Heuristics for Scheduling Parameter Sweep Application in Grid environments M1 Kenji KANEDA (SIG-PDS) – PowerPoint PPT presentation

Number of Views:80
Avg rating:3.0/5.0
Slides: 36
Provided by: kenj53
Category:

less

Transcript and Presenter's Notes

Title: Heuristics for Scheduling Parameter Sweep Application in Grid environments


1
Heuristics for Scheduling Parameter Sweep
Application in Grid environments
  • M1
  • Kenji KANEDA (SIG-PDS)

2
Outline of Presentation
  • Background Overview of AppLeS
  • (Task, Network) Model
  • Scheduling
  • Simulation Experiments
  • Related Work
  • Summary

3
Background (?????)
  • ?????????????????????????????????
  • ?????????????????
  • ????Firewall?Private Network??????????????????????
    ?
  • ??????????????????????????????????????

4
Background
  • ??????????????????????
  • Software DSM
  • MPI
  • parameter sweep application
  • embarrassing parallel
  • ??????????????????????

5
AppLeS (Application Level Scheduling)
  • parameter sweep application?????????
  • ??
  • ???parameter spaces
  • ????????
  • ????????????????
  • ????????????

6
AppLeS?????????
?? Server
Information Service
output
check
input
input
User
7
Outline of Presentation
  • Background Overview of AppLeS
  • (Task, Network) Model
  • Scheduling
  • Simulation Experiments
  • Related Work
  • Summary Future Work

8
Network Model (1/2)
  • network topology
  • k???????????????????
  • ????
  • ????????????????????
  • ??????????????
  • ??????????????????????????
  • ???????????????????(????????)

9
Network Model (2/2)
10
Task Model
  • ?????n?????
  • ???????????
  • ??????????????????
  • ??????????????????????????????????
  • ?????????
  • ????????????????????????
  • ????????????
  • MCell (micro-physiology application)
  • 3-D Monte-Calro simulation???
  • ?????????MBytes

11
Outline of Presentation
  • Background Overview of AppLeS
  • (Task, Network) Model
  • Scheduling
  • Simulation Experiments
  • Related Work
  • Summary

12
Scheduling
  • ??
  • makespan??????????????????
  • makespan ???????????????????????????????????????
    ???????????
  • adaptive?????????
  • ?????????????
  • ??
  • ????schedule()????????

13
schedule()
  1. ??schedule()???????????
  2. ???????????????????????
  3. ???????????????T???
  4. T??????Heuristic??????

Gantt Chart??????????????????????????
14
??schedule()????????
  • ??schedule()???????????
  • ????????????????????
  • ??????????????????
  • ?????????
  • ???
  • ?????
  • ????????????????????????????????????

15
????????? (1/2)
  • ??
  • ?????????????????
  • Gantt Chart??????????

16
????????? (2/2)
  • ?????????
  • NWS???????????????????Heuristic???
  • ?)???????????????
  • ??????????????????
  • ??????????????????????CPU load
  • ????????????????????CPU load
  • ??????????????

17
???????????????
  • schedule()???????????
  • ???schedule()???????????????
  • ?????????????????????????????
  • ????
  • random
  • maximizing data re-use
  • ???????????????????????????

18
????????????
  • ????????Heuristic??????????
  • Gantt Chart??????????????

Networks Links
Cluster 1
Cluster 2
Host 1.1
Host 1.2
Host 2.1
Host 2.2
Host 2.3
Link 1
Link 2
Link 1
input
output
19
Heuristics Algorithm (1/5) ?????
  • Hjkj th??????k th???
  • C(Ti,Hjk)???Ti????Hjk??????

??f(x)?????????x??
argmax???
20
Heuristics Algorithm (2/5) Min-min
Ti?????????????Hjk???
T??????????????Ts???
21
Heuristics Algorithm (3/5) Max-min
Ti?????????????Hjk???
T????????????????Ts???
22
Heuristics Algorithm (4/5) Sufferage
2????????????????????????Ti???
??????????????????????????
23
Heuristics Algorithm (5/5) XSufferage
???????????2???????????????????????Ti???
???????????????????????????
24
Simulation Environments
  • task
  • 1600 tasks (?????)
  • input unshared 10K file shared file
  • output 10K file
  • network topology
  • 5 clusters (6, 6, 8, 20, 20)
  • network bandwidth 6 600Kbytes/sec
  • interval of scheduling event 500sec

25
Simulation Result (1/2)
26
Simulation Result (2/2)
QoI performance estimation accuracy
27
Outline of Presentation
  • Background Overview of AppLeS
  • (Task, Network) Model
  • Scheduling
  • Simulation Experiments
  • Related Work
  • Summary

28
Related Work Condor(1/2)
  • ??
  • _at_University of Wisconsin-Madison
  • High Throughput Computing????
  • ???????
  • ??
  • ClassAd
  • ???????????????????????
  • OS, Memory, etc.
  • ???????????????????????????????

29
Related Work Condor (2/2)
  • ??
  • Remote System Call
  • remote host??local host???????????????
  • open()??????????????
  • checkpoint
  • ??????????????????????
  • fault tolerant
  • preemption
  • stack, process???????????????????
  • Fork(), kernel thread????

30
Related Work Nimrod
  • ??
  • _at_Monash University
  • Parameter sweep application?????
  • ??
  • Deadline scheduling
  • ????????????????????????????????????????????????
  • GUI interface ???

31
Related Work Bricks
  • ??
  • _at_Tokyo titech University
  • ????????????????????????
  • ??
  • Deadline Schedule?????????
  • Load Correction??????
  • Fallback??????

32
Related Work Scheduling on Heterogeneous
Environments
  • ???????????
  • regular graph
  • irregular graph
  • ??????? (???)
  • ????????
  • ??????????
  • ???????DAG
  • Task migration
  • ???????????
  • Throughput
  • round trip time

33
Outline of Presentation
  • Background Overview of AppLeS
  • (Task, Network) Model
  • Scheduling
  • Simulation Experiments
  • Related Work
  • Summary

34
Summary
  • AppLeS
  • parameter sweep application??????????
  • ?????Heuristics Algorithm???
  • ??????????????
  • XSufferage??????????????
  • ????????????????????
  • QoI??????????

35
References
  • Heuristics for Scheduling Parameter Sweep
    applications in Grid environments
  • Henri Casanova, Arnaud Legrand, Dmitrii
    Zagorodnov and Francine Berman
  • (HCW'2000)
  • Using Simulation to Evaluate Scheduling
    Heuristics for a Class of Applications in Grid
    Environments
  • Francine Berman, Henri Casanova, Dmitrii
    Zagorodnov and Arnaud legrand
  • Research Report
Write a Comment
User Comments (0)
About PowerShow.com