updating the Linux kernel

David Lombard david.lombard at mscsoftware.com
Mon Jun 12 10:56:18 PDT 2000


"Robert G. Brown" wrote:
> 
[deletia]
> 
> I would therefore NOT make auto-migration, robustness under node failure
> or checkpointing a "required feature of well-written parallel programs",
> or even the much weaker "advised/nice/suggested feature".  Sometimes
> this sort of bet-hedging is appropriate.  Sometimes it is essential.
> Sometimes it would be idiotic.  Only the programmer/designer, with a
> clear picture of the scaling properties and probabilities of failure for
> their own application and cluster hardware, is in a position to
> determine which category >>their<< particular application/cluster falls
> into.
> 

AGREED!

-- 
David N. Lombard
MSC.Software




More information about the Beowulf mailing list