Along the same lines, is there a way to run multiple UMLs in a system
_without_ guaranteeing them a slot of memory ?
The reason I ask is, let's say you have a machine with 2gigs -
realistically, looking at people running their name servers and web
servers and screening bitchx, etc., nobody ever uses their entire 128 megs
of ram (or 256 or whatever) that they are allocated. If you just played
fast and loose, I'll bet you could oversell, so to speak, the memory by at
Is there a way to do this ? I know that if I do not specify a memory
setting on the command line for a UML kernel, it just defaults to 64M (i
think), so that is not the way....
So basically, how do I oversell memory in my machine by not giving each
system a memory ceiling (and the flip-side to this is that the UMLs do not
allocate any more memory than they need for their current processes...)
On Thu, 11 Jul 2002, Cees de Groot wrote:
> Pavel Mihaylov wrote:
> > Hi,
> > I need to run several UMLs at the same time but each of them should have
> > access to a guaranteed part of the CPU power. It will be best to control
> > CPU access by percents, i.e. this UML will have 5% guaranteed and that
> > will have 10%.
> It'd be easiest if the host Linux system would support this, but fat
> chance. Personally, I would monitor CPU usage (read the relevant
> /proc/<pid>/ stuff every 30 seconds or so) and renice UML's down that
> were over quota. The exact algorithms depend on what level of
> granularity you put 5% on, of course (at some level of granularity, the
> time slice, any process will have 100% CPU available, so you need to
> move up 'high' enough to have this make sense; a second at least).
> Maybe there is some kernel extension that does this sort of scheduling;
> if not, it might be an interesting excercise to build one - it will have
> lots of value for UML users.
> This sf.net email is sponsored by:ThinkGeek
> PC Mods, Computing goodies, cases & more
> User-mode-linux-user mailing list