partitioning HD for use of swap & for booting

Nordwall, Douglas J Nordwall at pnl.gov
Fri Jun 1 09:37:04 PDT 2001


I guess the question I have got to ask is if it is still practical to run with
2xmemory when you start reaching up into large memory? I assume this is only if
you ever hit swap...

> -----Original Message-----
> From: Bogdan Costescu [mailto:bogdan.costescu at iwr.uni-heidelberg.de]
> Sent: Friday, June 01, 2001 8:59 AM
> To: Robert G. Brown
> Cc: Kent Heyward; Beowulf List
> Subject: Re: partitioning HD for use of swap & for booting
> 
> 
> On Fri, 1 Jun 2001, Robert G. Brown wrote:
> 
> > I've heard that 2x memory is the MINIMUM recommended for the
> > 2.4 kernels for performance reasons.
> 
> This was specifically mentioned when 2.4.0 was launched as it's an
> important difference w.r.t. 2.2 line.
> 
> > I'd even speculate that a 2.4 kernel
> > with lots of memory but small swapspace is a better 
> performer than a 2.4
> > kernel with barely enough memory but plenty of swapspace.
> 
> No, a 2.4 kernel with small swap space will perform poorly 
> especially with
> latest (2.4.4, 2.4.5) kernels. There was/is a discussion on lkml about
> that; some people say that from 2.4.0, with every new release, the
> behaviour got worse. Some even say that 2.4.x should not be considered
> "stable series" because of the VM problems.
> 
> > Linux has a very decent memory management system, and generally ...
> 
> I guess the more appropiate form is "should have".
> 
> Please don't take me as a VM expert. I'm subscribed to lkml 
> and read most
> of the messages, that's the source for all the above info. However, as
> Linux workstation user, I've noticed a clear degradation 
> between 2.2 and
> 2.4, even some stability problems.
> (To get back on the beowulf track) Until 2.4 behaves at least 
> as good as
> 2.2 in all aspects, I'm only running 2.2 on the clusters that 
> I administer.
> 
> Back on the swap track: we're on the beowulf list and this 
> question should
> be in fact related to usage of swap in clusters. I always found a bad
> ideea running a parallel application that doesn't fit in RAM (well,
> probably a very coarse grained one might work OK). So I'd 
> second Robert's
> suggestion to buy more RAM if possible.
> 
> Sincerely,
> 
> Bogdan Costescu
> 
> IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
> Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
> Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
> E-mail: Bogdan.Costescu at IWR.Uni-Heidelberg.De
> 
> 
> 
> 
> _______________________________________________
> 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