WestGrid Seminar Series Copyright © 2006 University of Alberta. All rights reserved Integrating...

75
Copyright © 2006 University of Alberta. All rights reserved WestGrid Seminar Series Integrating Gridstore Into The Job Submission Process With GSUB Edmund Sumbar University of Alberta

Transcript of WestGrid Seminar Series Copyright © 2006 University of Alberta. All rights reserved Integrating...

Copyright © 2006 University of Alberta. All rights reserved

WestGrid Seminar Series

Integrating GridstoreInto The Job Submission

ProcessWith GSUB

Edmund Sumbar

University of Alberta

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 2

GSUB

Gridstore…• close proximity to vizserver

• does not impose quotas

• has a large capacity

• performs backups

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 3

GSUB

time gcp gridstore:largefile .

glacier lattice *matrix nexus cortex

10MB 4 5 8 6 5

100MB 14 7 70 40 170

1GB 120 55 ~600 440 ~4000

Data collected January 11, 2006.

* Using scp, gcp not installed yet.

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 4

GSUB

Job submission protocol…• copy files from gridstore to execution site (gcp)

• submit job (qsub)

• wait for job to end (qstat)

• copy files from execution site to gridstore (gcp)

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 5

GSUB

desktop

user

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 6

GSUB

{execution nodes

desktop

user

head node

exec site

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 7

GSUB

{execution nodes

desktop

user

head node

exec site

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 8

GSUB

{execution nodes

desktop

user

head node

exec site

head node

home site

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 9

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 10

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

grid-proxy-init

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 11

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

grid-proxy-init

ssh

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 12

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh

gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 13

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh qsub

gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 14

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh qsub

gcp gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 15

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh qsub

gcp gc

p

1

2

3

4

5

6

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 16

GSUB

Becomes tedious/inefficient/error-prone

Automate process with scripts…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 17

GSUB

Consider gsub…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 18

GSUB

Consider gsub…• No changes are required to your existing Torque (PBS)

job scripts

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 19

GSUB

Consider gsub…• No changes are required to your existing Torque (PBS)

job scripts

• All qsub command-line options are supported

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 20

GSUB

{execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 21

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 22

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

grid-proxy-init

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 23

GSUB

ssh {execution nodes

desktop

user

head node

exec site

head node

home site

grid-proxy-init

ssh

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 24

GSUB

{execution nodes

head node

exec site

ssh

desktop

user

gridstore

head node

home site

grid-proxy-init

ssh gsub

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 25

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh gsub

gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 26

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh gsub

gcp

stage in

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 27

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh qsub

gcp gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 28

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh qsub

gcp gcp

stage out

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 29

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh gsub

gcp

gcp

1

2

3

4

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 30

GSUB

Demonstration…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 31

GSUB

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 32

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh gsub

gcp

gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 33

GSUB

ssh {execution nodes

desktop

user

head node

exec site

gridstore

head node

home site

grid-proxy-init

ssh gsub

gcp

gcp

blackhole(gridstore head node)

ssh• prepare input

• analyze output

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 34

GSUB

Submit jobs directly from blackhole…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 35

GSUB

desktop

user {execution nodes

head node

gridstore

head node

home site

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 36

GSUB

desktop

user {execution nodes

head node

gridstore

head node

home site

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 37

GSUB

desktop

user {execution nodes

head nodeblackhole

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 38

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 39

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

myproxy-get-delegation

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 40

GSUB

head nodeblackholedesktop

user {execution nodes

ssh

myproxy-get-delegation

gridstore

gsub

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 41

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

myproxy-get-delegation

gsub gsub

gridstore

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 42

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

myproxy-get-delegation

gsub gsub

gridstore

gcp

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 43

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

myproxy-get-delegation

gsub gsub

gridstore

gcp gc

p

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 44

GSUB

desktop

user {execution nodes

head nodeblackhole

ssh

myproxy-get-delegation

gsub gsub

gridstore

gcp

gcp

1

2

3

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 45

GSUB

Demonstration…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 46

GSUB

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 47

GSUB

Tools for monitoring and managing jobs• gstat

• gdel

• glog

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 48

GSUB

Demonstration…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 49

GSUB

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 50

GSUB

View WestGrid like a scaled-up execution

site…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 51

GSUB

desktop

user {execution nodes

head node

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 52

GSUB

desktop

user

{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 53

GSUB

desktop

user

{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 54

GSUB

desktop

user

{{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 55

GSUB

desktop

user

{{{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 56

GSUB

{desktop

user

{{{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 57

GSUB

{desktop

user

{{{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 58

GSUB

{desktop

user

{{{

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 59

GSUB

{desktop

user

{{{

blackhole

super head node

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 60

GSUB

{desktop

user

{{{

blackhole

super head node

myproxy

credential server

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 61

GSUB

{desktop

user

{{{

blackhole

super head node

myproxy

credential server

gridstore

file system

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 62

GSUB

{desktop

user

{{{

blackhole

super head node

myproxy

credential server

gridstore

file system

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 63

GSUB

{desktop

user

{{{

blackhole

super head node

myproxy

credential server

gridstore

file system

WestGrid

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 64

GSUB

DISCLAIMER

The opinions expressed herein are those of

the author and do not necessarily represent

the views of WestGrid management.

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 65

GSUB

Gsub implements programmable workflow

Specify actions to be executed at various

times (phases) during the life of a job• before stage in

• after stage in

• job start

• job end

• before stage out

• after stage out

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 66

GSUB

Workflow programmed as a case statement

case $phase in beforestagein ) exit ;; afterstagein ) exit ;; jobstarted ) exit ;; jobended ) exit ;; beforestageout ) exit ;; afterstageout ) if some condition is true; then resubmit fi exit ;;esac

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 67

GSUB

Predefined workflows• email

• prevent

• tidy

• autoresub

• exclusive

• aggressive

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 68

GSUB

No script modifications required

Specified at submission time• gsub jobscript=workflow

Cascade workflows• gsub jobscript=workflow1+workflow2

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 69

GSUB

Demonstration…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 70

GSUB

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 71

GSUB

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 72

GSUB

Gsub is unsuitable for…• very short jobs

• developing, testing, debugging programs

• staging very large or very many files

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 73

GSUB

Gsub is not officially accepted by WestGrid

Still needs testing in the real world…

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 74

GSUB

Gsub is not officially accepted by WestGrid

Still needs testing in the real world…

…any volunteers?

Copyright © 2006 University of Alberta. All rights reserved create deleteJanuary 25, 2006 75

GSUB

Questions…