thrashing

John Brookes johnb at quadrics.com
Thu Aug 28 17:22:13 EDT 2003


The test I would probably suggest to someone whose machine I had no access
to is 'swapoff -a'. It's not big and it's not clever, but largely removes
the need for value judgements: if it bombs in an OOM style, you were most
probably thrashing.

Just a thought.

Cheers,

John Brookes
Quadrics

> -----Original Message-----
> From: Glen Kaukola [mailto:glen at mail.cert.ucr.edu]
> Sent: 28 August 2003 19:19
> To: beowulf at beowulf.org
> Subject: thrashing
> 
> 
> Hi there,
> 
> So for our newest simulations, we're working with a different domain, 
> where each of our grid cells are much smaller, and so we're expecting 
> the runs to take about 4 times longer.  But actually they're taking 
> around 40 times longer.  I'm thinking this may have something 
> to do with 
> not having enough memory.  The problem with this theory is 
> that I'm not 
> really sure how to tell if my machines are thrashing.  On a desktop 
> machine I can tell no problem, as the disk starts going crazy and the 
> system pretty much grinds to a halt.  But on a machine up in 
> my server 
> room on which I don't have any gui and where it's too loud to 
> hear any 
> disk activity, I'm really not sure how to tell whether it's 
> thrashing or 
> not.  I mean, I can look at top, and free, and sar and everything 
> doesn't look much different than when the other simulations were 
> running, except for maybe 'sar -W', which is a little bit higher.  
> Anyway, if someone could help me out with a way to determine 
> without a 
> doubt if my machines are thrashing or not, then I'd greatly 
> appriciate it.
> 
> Thanks for your time,
> Glen Kaukola
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) 
> visit http://www.beowulf.org/mailman/listinfo/beowulf
> 
_______________________________________________
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