Re: [uml-devel] Re: [uml-user] production

[Matthew B. <ma...@by...>]

On Friday 06 December 2002 21:57, Steve Schnepp wrote:
> On Sat, Dec 07, 2002 at 09:29:16AM +1300, Cameron Kerr wrote:
> > You want to let the host kernel cache, that's a Good Thing. If the host
> > needs more memory it will simply reduce the size of the filesystem
> > cache.
>
> And here I come again... I mean, it seems that SKAS is a Wonderful
> Thing, since free mem in the UML is also free mem on the host, unlike
> the POTM [ Plain Old TT Mode ;-p ] which use all the mem at once. So
> now, cutting the throat to pagecaching [ mostly inode caching ] means
> more "free" mem on the UML, and that way more free mem on the host which
> could happily cache thing with better hints than UMLs.

Steve, did you ever run any tests as to whether forcing O_DIRECT to open() 
calls inside a UML resulted in better performance?  ISTR you saying you were 
going to when this was discussed last.  As I said before I'm not sure why you 
wouldn't get better performance through being able to share a disc cache.

-- 
Matthew Bloch         Bytemark Computer Consulting Limited
                                http://www.bytemark.co.uk/
                                  tel. +44 (0) 8707 455026