Re: [uml-user] Hostfs/NFS/UML weirdness

[Jeff D. <jd...@ka...>]

sf...@ih... said:
> I would assume a box with 4 GB RAM could run 128 UMLs, if each was
> using 32 megs of RAM, which a small distro with Apache, sendmail and
> named could easily fit into. Any reason that my thought is incorrect? 

That seems reasonable to me.

And I think you could improve on that by making all the UMLs COW from the
same root_fs and mounting them -o sync.  I think this will reduce UML memory
consumption by eliminating the file caching from the UMLs, which would 
otherwise be keeping their own separate copies of the same data, and move 
all the caching to the host, where all the UMLs can share it.  And there 
shouldn't be much of a performance hit if it's mostly in the host's cache.

If anyone wants to try that with large numbers of UMLs, I'd be interested in
the results.

				Jeff