Re: [PATCH] nfs: fix congestion control

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Fri, 2007-01-19 at 18:57 +0100, Peter Zijlstra wrote:
> On Fri, 2007-01-19 at 09:20 -0800, Christoph Lameter wrote:
> > On Fri, 19 Jan 2007, Peter Zijlstra wrote:
> > 
> > > +	/*
> > > +	 * NFS congestion size, scale with available memory.
> > > +	 *
> > 
> > Well this all depends on the memory available to the running process.
> > If the process is just allowed to allocate from a subset of memory 
> > (cpusets) then this may need to be lower.
> > 
> > > +	 *  64MB:    8192k
> > > +	 * 128MB:   11585k
> > > +	 * 256MB:   16384k
> > > +	 * 512MB:   23170k
> > > +	 *   1GB:   32768k
> > > +	 *   2GB:   46340k
> > > +	 *   4GB:   65536k
> > > +	 *   8GB:   92681k
> > > +	 *  16GB:  131072k
> > 
> > Hmmm... lets say we have the worst case of an 8TB IA64 system with 1k 
> > nodes of 8G each.
> 
> Eeuh, right. Glad to have you around to remind how puny my boxens
> are :-)
> 
> >  On Ia64 the number of pages is 8TB/16KB pagesize = 512 
> > million pages. Thus nfs_congestion_size is 724064 pages which is 
> > 11.1Gbytes?
> > 
> > If we now restrict a cpuset to a single node then have a 
> > nfs_congestion_size of 11.1G vs an available memory on a node of 8G.
> 
> Right, perhaps cap this to a max of 256M. That would allow 128 2M RPC
> transfers, much more would not be needed I guess. Trond?

That would be good as a default, but I've been thinking that we could
perhaps also add a sysctl in /proc/sys/fs/nfs in order to make it a
tunable?

Cheers,
  Trond

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/