[Foxtrot] Re: MPQS matrix reduction in parallel (fwd)
Alan Lawrence
acl33@hermes.cam.ac.uk
Sat, 10 Feb 2001 18:43:54 +0000 (GMT)
As expected, yes :-(
We can't really go against what he is telling us to do, which is a
shame; I rather liked the all-processing-performed-on-client
approach. OTOH, if we do decide to "have a go" and get something that
works, that'll put us closer to the cutting edge of research (right _on_
the cutting edge, one might say) than any other group. Prize anyone? :-)
Alan "head in the clouds" Lawrence
On Sat, 10 Feb 2001, Michael Pinna wrote:
> Here's the reply... As I suspected he's telling us not to bother with the
> interesting solution.
>
> Mike
>
> --
> "...like the liberated debater toasting the bowl."
> --Anon
>
> ---------- Forwarded message ----------
> Date: Sat, 10 Feb 2001 16:06:38 +0000
> From: Ross Anderson <Ross.Anderson@cl.cam.ac.uk>
> To: Michael Pinna <map36@cam.ac.uk>
> Cc: ross.anderson@cl.cam.ac.uk, Ross.Anderson@cl.cam.ac.uk
> Subject: Re: (group project foxtrot) MPQS matrix reduction in parallel
>
> For the size of problem you're likely to be working with, you
> should just accept that the matrix reduction will be done on
> one machine, even if it limits the size of the number you can
> factor to a hundred digits or so.
>
> There are ways of distributing the matrix reduction, but they
> are still in the category of `rocket science' rather than
> undergraduate project material. The world expert on them is
> Paul Leyland at Microsoft Research across the road; by all
> means talk to him if you wish, but I expect he will discourage
> you from trying to implement complex and poorly-understood
> algorithms (block Lanczos) in a time-limited project. Even if
> you decide to `have a go', for goodness' sake leave it as an
> optional extra that you can tackle once you have a project
> that otherwise works and will get you your ticks,
>
> Ross Anderson
>
>
>
> _______________________________________________
> Foxtrot mailing list
> Foxtrot@srcf.ucam.org
> http://www.srcf.ucam.org/mailman/listinfo/foxtrot
>