In June 2004 I stopped doing NFS as a distributed project and switched to
using the 15 nodes beowulf cluster Pingo at Risø National Laboratory, where I have
been given access to run jobs when it's idle.
In the beginning of November 2004 I moved the work to the 240 nodes cluster
Mary at the
This cluster has recently stopped being in testing and has started
production, so I can no longer use it.
This is the description of the old system
These pages are about my client/server system for SNFS factoring,
it was originally meant for Conrad Curry's NFSNET effort, to help with the
assignment and distribution of ranges for sieving, but since the apparent
death of that project and Conrad's disappearance I've switched it to handing
out ranges from private projects such as
38,91+, finished on my birthday:
Used to factor numbers from Andy Steward's
part of the Cyclotomic
Number Project. the description below is old, the system isn't running at the moment
The current method for handing out ranges is to first check if there are any
private projects for the account, and hand out a range for that if there is.
If not, it selects the project with the largest expected memory use which is
smaller than the one given in the request, and hands out the first range for
The idea behind this is that the maximum memory requirement is configured on
each client machine, so an account with multiple machines will get the
"largest" assignments available for each machine.
The size of the range given out is decided by the time it took that machine
to return the results the last couple of times, and currently aims at a
range that takes one day to finish.