[Beowulf] To the bandwidth junkies

Håkon Bugge hbugge at platform.com
Thu Jan 29 16:52:45 PST 2009


On Jan 30, 2009, at 1:42 , Joshua mora acosta wrote:

> In the first loop k should start with 0 instead of 1...
> Then you divide by NTIMES.

No, the first k loop starts intentionally at 1, to avoid distortion of  
avg and max due to initial page faults in the first pass. I concluded  
what the right fix is, and, as Tom Elken points out in a private mail,  
the fix has already been implemented in other version of stream, see http://www.cs.virginia.edu/stream/FTP/Code/stream.c



Thanks, Håkon


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scyld.com/pipermail/beowulf/attachments/20090130/4dc12c0e/attachment.html


More information about the Beowulf mailing list