Outreach

1
Outreach
Distributed computing
Anthrax research project
2
Why does ISIS need faster processing?

• Many compute intensive programs and problems
• Many applications which could be compute
  intensive
• Large range of application domains
• Simulations (to plan experiments)
• Analysing results
• Designing new instruments
• Visualisation

3
Spare Cycles Concept

• Typical PC CPU usage is about 10
• Most PCs not used at all after 5pm
• Even with heavily used (Outlook, Word, IE)
  PCs, the CPU is still grossly underutilised
• Everyone wants a fast PC!
• Can we use (steal?) unused CPU cycles
  from the plethora of PCs?

4
Our United Devices Grid-MP System

• Nerve centre of the grid
• Two servers controlling submission and
  distribution of jobs
• Compute resources on the grid
• Currently 200 desktop Windows PCs of varying
  power.
• UD software runs submitted jobs silently on PCs
  PC useris not aware his / her PC isbeing used.

5
Visual introduction to the grid
Servers
Desktop
Client PCs
6
Suitable / Unsuitable Applications

• CPU Intensive
• Low to moderate memory use
• Not too much file output
• Coarse grained
• Command line / batch driven
• Licensing issues?

7
How to write Grid Programs

• Fairly easy to write
• Interface to grid via Web Services
• So far used C, Java, Perl, Fortran, C

• Think about how to split your data and merge
  results
• Wrap and upload your executable
• Write the application service
• Pre and Post processing
• Use the Grid

8
Ray tracing

• Photorealistic image creation by ray tracing
• Very compute intensive
• Several days, weeks, even months per image

9
DASH Crystal structure determination

• Multi-solution SA (McBride/Johnston/Fernandes
  PhDs)
• Execute a number of independent SA runs (say 25)
  pick the best one

10
Not just faster, but better

• Rule of thumb simple compounds solve quickly
  and reliably, complex compounds solve slowly
  and with less reliability

Hydrochlorothiazide 20s per SA run 100 success
rate
Verapamil hydrochloride 90 minutes per SA run 5
success rate
11
Algorithm sweet spot
Good
Calculations embody ca. 6 months of CPU time.
On our current grid, runs would be completed in
ca. 24 hours.
Poor
12
Monte Carlo Speed-up Ideas

• Two scenarios
• Single large simulation run
• Split the neutrons into smaller numbers and
  execute separately
• Merge results in some way
• Many smaller runs
• Parameter scan

13
Instrument Simulation

• Simulation of TS2 instruments increasingly
  important
• Simulation of new HRPD guide
• Vitess
• McStas

14
VITESS Some Results
176 hours 59 hours
6hrs 20mins
15
McStas Splitting It

• Different executable for every run
• Executable must be uploaded at run time
• Split n into chunks
• or run many instances (parameter scan)
• Create data ( executable) packages
• Upload packages

16
Advantages
CPU Power Tap into existing power instantly Power
of the Grid increases with time Investment Excelle
nt example of a Gershon-type saving i.e. making
better use of existing resources Environment Machi
nes are burning electricity, so why not use them?
17
The future
Increased CPU Power Deploy client software
outside of ISIS We have an unlimited CCLRC-wide
client licence agreement Increased number of
applications The more applications that are
enabled, the more value we will get from the
grid Increased ease-of-use The easier it is to
use the grid, the more likely it is to be used
18
The problems
Real Not that many! Occasional drug interaction
type problem Imagined People are, by nature,
suspicious of the idea Whenever a machine goes
slow, the grid is blamed The answer A top-down
approach i.e. convince system managers Constantly
monitor the system Be flexible in scheduling
24/7 is possible, but is it desirable?