Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: wallclock time versus cpu time

Author: Robert Hyatt

Date: 10:53:18 09/04/02

Go up one level in this thread


On September 04, 2002 at 13:27:13, Vincent Diepeveen wrote:

>On September 04, 2002 at 12:43:37, Robert Hyatt wrote:
>
>In short your new statement here says you used a
>cpu time instead of wallclock time?

No it didn't.  And I am about ready to give up.

The speedups are computed from wallclock time _only_.  The times
used to produce them were wall-clock.

The only instance of cpu time was a bad reference.  As I said,
very _clearly_ each cpu very carefully kept up with the amount of
time it spent searching, and the amount of time it spent "spinning
waiting on work or locks."  That was _also_ wall-clock time as that
is all we used.  CPU time on the cray was a problem in that all the
threads got their times added together which made it not so useful,
other that to compute a cpu% number we used as a warning.  When it
dropped below (say) 1550% we knew someone else had snuck in and was
using our machine.  this happened several times during the career of
Cray Blitz.

But _no_ cpu time in any of those three tables.

Can I be more specific than that?


>
>If you reserve N cpu's at a supercomputer, you get N cpu's
>at a supercomputer; this was not the case in your research,
>so if i inform at the organisation whether they gave you
>not n cpu's to do n cpu's test i get the answer: "he just
>had to run at it and had to be lucky when he got cpu time,
>but his processes were not killed".



What are you talking about?  When I ran the 16 cpu game, I was using
a 16 cpu C90 with no other users whatsoever.  When I ran the 2,4 and 8
cpu tests, I used an 8 cpu C90 with no other users.  When I ran the one
cpu test, then I ran with others, but I had one cpu guaranteed to me
because I ran at a real-time priority and no one else could.  So I
don't know what you are talking about above...  Certainly not about
anything I did...

>
>That is the answer i get from the sponsor?



No idea.  I can't read minds or predict the future.  but it has nothing
to do with the testing I did..  so it really doesn't matter to me at all.

Why don't you just tell your sponsor you are getting a 17X speedup on 16
cpus and get it over with?  That seems to be _the_ issue here.  Not truth.
Not accuracy...  so just go for it and have fun.






This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.