[Gllug] Recommended distro

Christian Smith csmith at micromuse.com
Tue Dec 14 17:44:33 UTC 2004


On Tue, 14 Dec 2004, John Hearns wrote:

>On Mon, 2004-12-13 at 17:14 +0000, Christian Smith wrote:
>
>> In a commercial environment, I'd have thought reliability and support are
>> major contributors to whether to choose PCs or SUN kit.
>>
>> For CPU clusters, yes, go for the cheap, throw away PCs. Reliability is
>> not of as great concern as performance.
>Cough.
>Reliability is important - for instance using ECC memory,
>well tried components and having a good burn-in procedure,
>and paying attention to the cooling.


Sorry, I worded that badly. Individual availability is not important. Data
reliability is paramount, of course.


>
>I agree though that it doesn't make sense to have, for example,
>redundant PSUs or mirrored disks on compute nodes.
>If they go down, there are other nodes in the cluster.


Yeah, that's more along the lines I was thinking of. Any server should
have ECC memory.

Reminds me of an interesting story from one of the POWER articles by an
IBM guy. They had to change their CPU test to include a benchmark, as
faulty CPUs were retrying instructions till they succeeded such was the
reliability functionality of the latest POWER processors. Bollocks to PCs
and SUNs. Given that, I'd go for a eServer p5 series machine:)


>
>You CAN arrange checkpointing of your code, so you can restart on
>other nodes. But that's a user-space task. Grin.
>
>Its common to have redundant features on a compute cluster master node.
>
>And support is important of course - thats what clustering companies
>provide.
>
>
>ps. I shall redeem myself at the next GLLUG or LONIX by drinking beer
>and discussing silly things.

For a change...


Christian

-- 
    /"\
    \ /    ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL
     X                           - AGAINST MS ATTACHMENTS
    / \
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list