1. No category

Build it and they will come: How to get
non-physicists to use your Grid
Stuart Purdie
Your university or
experiment logo here
Overview
• Setting the scene
– With pictures
• Some specific case studies
– i.e. ones that worked...
• Where to go from here
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
2
Mauna Kea
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
3
Mauna Kea Observatory
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
4
Mauna Kea
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
5
User start here?
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
6
User start here!
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
7
How we see it
B
A
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
8
How they see it
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
9
Big Questions
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
10
Big Questions
“What's ssh?”
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
11
Picture of very hard mountain here
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
12
Picture of very hard mountain here
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
13
Things users find difficult
• Mental model of how the Grid works
• After that:
–
–
–
–
–
–
Batch computing
Job submission
Job monitoring
“Where's my output?”
Not killing a site
Meta-scheduling
• So … just about everything
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
14
Things users find difficult
• Mental model of how the Grid works
• After that:
–
–
–
–
–
–
Batch computing
Job submission
Job monitoring
“Where's my output?”
Not killing a site
Meta-sceduling
• So … just about everything
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
15
Some case studies
•
•
•
•
Electrical Engineering
Optical Engineering
Solid State Physics
Statistics
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
16
NanoCMOS
• Electrical Engineers
– Running their own code
• Dates back to beginning of the the cluster at
Glasgow
• Was always meant to be a Grid Project
• Used shared areas a lot
– Not quite best fit for Grid systems
– AFS as a replacement
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
17
Optical Engineering
• Driven by Lumerical
• Commercial software; 10 user licence
granted by Lumerical
– Available to UK opto-engineering
community
• Wrote custom driver scripts to hide the input
and output staging
• Can use MPI; often does
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
18
Solid State Physics
•
•
•
•
Primarily multi core MPI jobs
CASTEP
Lots of CASTEP
Interesting scaling behaviour – nodes for peak
performance for our GigE coupled network
same as on HECTOR
– (HECTOR still a wee bit faster, of course)
• And more CASTEP
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
19
Statistics
• R package
– (Also in use by other VO's: DAMES)
• Interesting quirk on R Libraries, and working
out how to handle them
• Users were inexperienced with batch
computing
– But picked it up quickly
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
20
Overall points from case studies
• Something to drive adoption
– Normally use of backfill for free
• Use of commercial software licences might
be an avenue worth exploring
– Small number of licences resulting in wide
net
• User education is always more tricky than
expected
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
21
Not users yet
Third (and final!) section
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
22
Some things that are happening
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
23
Some things that are happening
• Software that wasn't designed for the Grid
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
24
Some things that are happening
• Software that wasn't designed for the Grid
• Software that wasn't designed for Batch
computing
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
25
Some things that are happening
• Software that wasn't designed for the Grid
• Software that wasn't designed for Batch
computing
• Software that wasn't designed
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
26
Things the Grid won't do well
• But is done
• Software that assumes a GUI
• Software will multithread up to the 'whole
node' rather than a set list of processes. Not
MPI – must be same machine.
• Software that assumes it's run on a desktop –
and can monopolise it
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
27
Things the Grid won't do
• But people want / expect
• HTTP stage-in support
• CPUTime / WallTime support for Job
submission
• MPI jobs
– Fabric matters – InfiniBand is different
from Gig Ethernet!
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
28
NGS
• NGS affiliate
– Support ngs VO
– And some basic software (that's probably
already installed)
• Not too hard
• Decent back fill
• VO's work better the more sites they are
enabled at
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
29
NGS VO usage
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
30
26/08/10
Non-physicists on your Grid
Your university or
experiment logo here
31