The UCL Condor Pool Experience

1
The UCL Condor Pool Experience
Environment from the Molecular Level A NERC
eScience testbed project

• John Brodholt1, Paul Wilson3, Wolfgang Emmerich2
  and Clovis Chapman2.
• 1. Department of Earth Sciences, University
  College London, Gower Street, London WC1E 6BT, UK
  
• 2. Department of Computer Science, University
  College London, Gower Street, London WC1E 6BT, UK
  
• 3. Anvil Software, London, UK.

2
Environment from the Molecular Level A NERC
eScience testbed project
The UCL Condor Pool
Approximately 946 Windows machines (yesterday) 1
to 2.4 GHtz Intel processors 256 to 512 MBytes
memory (a few are more) They are in open
access student cluster rooms PCs are all thin
client WTS machines with network bootable
operating systems. (Citrix/Bpbatch - hit spacebar
to upload new operating systems image) The pool
is very simple one manager, one submit machine
(via ssh).
3
Environment from the Molecular Level A NERC
eScience testbed project

• Q. Why was it someone from an Earth Science Dept.
  who got it going?
• Because three years ago, the eScience grants call
  made me look up the Grid on the web and by
  chance I came across the Condor web site.
• I also happened to know how Information Systems
  at UCL managed their student PCs.
• Persuaded the Director of UCLs Education and
  Information Systems Division that I could put it
  in our eMinerals grant (I think he assumed it
  wouldnt get funded).

4
Environment from the Molecular Level A NERC
eScience testbed project
Key Political Issues
Even though the Director of EISD had agreed for
us to put it in the grant, we had to convince
Information Systems themselves. Numerous
meetings . IS produced a five page document
outlining what they thought their policy on a
large Condor cluster would be i.e. the primary
purpose of the student cluster rooms must not be
compromised. Nor should IS staff use their time
on the project etc. Needed testing (one
cluster, then one image type). Perhaps the key
moment was when the UCL presented its eScience
projects to Tony Hey and the UCL Provost.
5
Environment from the Molecular Level A NERC
eScience testbed project
Timescale
Desktop - June 2002 (2 nodes) Earth Science
Student Cluster Room - Oct 2002 (18
nodes) Physics Department (one WTS image) Jan
2003 (150 nodes) Campus October 16th 2003 (930
nodes) 1 millionth hour of CPU April 2nd,
2004 This matched exactly the timescale we
outlined in the eMinerals grant
6
Environment from the Molecular Level A NERC
eScience testbed project
Other Issues
Difficult to persuade the scientists to get
involved for just a few machines.
Some needed to compile their codes for Windows
machines Its simple, just convert them to
Java .. Wolfgang Emmerich, 2002!
Our central manager died a few times when a user
submitted a few thousand jobs all at the same
time (took 24 hours to repair disk with fsck).
Now have a manager and a submit machine.
Students will do anything to reserve a machine
steal the mouse, put out of order signs on them,
and UNPLUG them. Also, IS themselves briefly
turn machines in some clusters off in order to
clear the room. This restricts the length of job.
7
Environment from the Molecular Level A NERC
eScience testbed project
8
Spikes in user demand
Environment from the Molecular Level A NERC
eScience testbed project

• Not many users
• Most are using simple schemes to produce lots of
  initial input files and send off to pool. Get
  results back and spend a long time processing
  them/extracting data/planning next set of inputs.

9
Existing e-science technology
User Input Structural model Si/Al, cation types,
H2O etc.
Distributed resources (Condor pools etc.)
Distributed Computing Portal
Model/Configuration Generator
Analysis(geometry, energy, fit)
Improve generation / modelstrategy
User Input Diffraction data, chemical
analysis,building units Si/Al, cation types,
H2O etc.
Drip feeding and interactive steering of a Condor
pool using relational databases Dewi Lewis, Rosie
Coates and Sam French UCL Chemistry / RI
10
THE Science.1. Simulation of pollutants in the
environment Binding of heavy metals and
organic molecules in soils. 2. Studies of
materials for long-term nuclear waste
encapsulation Radiocactive waste leaching
through ceramic storage media. 3. Studies of
weathering and scaling Mineral/water
interface simulations, e.g oil well
scaling.also4. The Earths core and
mantleMany codesDL-POLY, GULP, METADISE,
CRYSTAL, CASTEP, SIESTA,
Environment from the Molecular Level A NERC
eScience testbed project
11
Now what?
Environment from the Molecular Level A NERC
eScience testbed project
Expand pool to include staff WTS machines 1500
machines (received 3 page email from IS - who
owns them?).
UCL Staff machines at hospitals ???? machines.

Federate with other pools hopefully make it
more flexible smooth spikes in demand.