Reliability, WAS: 0,000,000 for cluster computing

John Brookes johnb at quadrics.com
Tue Nov 12 11:40:24 EST 2002



John Brookes
Quadrics
T: +44 (0)117 9155500
F: +44 (0)117 9075395
E: johnb at quadrics.com
W3: www.quadrics.com


> -----Original Message-----
> From: John Brookes 
> Sent: 12 November 2002 16:38
> To: 'Alan Scheinine'; John Brookes
> Subject: Reliability, WAS: 0,000,000 for cluster computing
> 
> 
> Nah. I agree that a user shouldn't be annoyed with the system 
> or personnel if a long-running job dies due to node-failure 
> (or any other sufficiently justifiable reason), but they have 
> a right to be irritated. Based on a (very common) premise of 
> paying for storage, the case for writing intermediate results 
> amounting to high-end gigs (or teras) of intermediate results 
> is often weak - even in comparison to losing results due to 
> failure. The point is, "Do I rewrite the code to record 
> intermediate results, or can I reasonably expect the system 
> to complete most jobs without failure?" What's going to be 
> more expensive, basically. Buyers should, of course, ask 
> these questions of themselves/vendor, but there should be a 
> sensible metric. In the case of flops, one may refer to 
> benchmarks or sample runs. Obviously such b/m results only 
> give a guide and can be misleading if the tests are 
> inappropriate, but they give you an idea, at least. 
> 
> John Brookes
> Quadrics
> T: +44 (0)117 9155500
> F: +44 (0)117 9075395
> E: johnb at quadrics.com
> W3: www.quadrics.com
> 
> 
> > -----Original Message-----
> > From: Alan Scheinine [mailto:scheinin at crs4.it]
> > Sent: 12 November 2002 16:14
> > To: johnb at quadrics.com
> > Subject: RE: 0,000,000 for cluster computing
> > 
> > 
> > This message uses a character set that is not supported by 
> > the Internet Service.  To view the original message content,  
> > open the attached message. If the text doesn't display 
> > correctly, save the attachment to disk, and then open it 
> > using a viewer that can display the original character set. 
> > 
> 
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list