IWAMOTO Toshihiro wrote:

> At Thu, 04 Jan 2001 21:43:36 +0100,
> Lars Heidieker wrote:
> >
> > Hi,
> >
> > it might be that doing memory pageouts --- done to increase the vnode
> > cache --- hurt the io,
> > and breaking the clustering by that.
> > does the same happen if there is plenty of free mem?
>
> Before sending the previous mail, I also tested with a 1.5K kernel
> with 100MB of bufcache.  Interestingly, "tar zxf" sometimes ran slower
> than it did with 50MB bufcache.  So it might be the case that too much
> of free mem hurts performance.
> (Beware -- this might be a false indication.  More careful tests are
> needed to conclude something.)
>
> --
> IWAMOTO Toshihiro

Hi,

yes that might be true, but I thought of having free mem so that the vnode
cache does not cause pageouts and not a very big bufcache which is not
very usefull on a UBC System anyway.
Its probably the LRU approximation on the bufferheads that cause a slow
down if their count is very high --- Just a thought.

lars