From: Jeff Dike <jdike@ka...> - 2001-08-30 14:46:13
> Using the small debian root_fs I achieved 31 umls running
> simultaneously okay. More than that seemed to give paging problems/
From UML? What were they exactly?
> I also tried the above setup but using the RH7.1 pristine root_fs and
> only managed to get around 10 umls running simultaneously without
The RH7.1 fs is a lot more memory-hungry than the Debian fs.
> Is there any limit caused by the maximum number of Linux virtual
> consoles that can exist??? I've read (man console) that Linux can
> have a maximum of around 63 (or 64?) virtual consoles.
It'd be nice to know what exactly it depends on. Did you play any tricks
to get that number up? I assume you COWed the filesystems. One other thing
you might try is having the UMLs mount their root filesystems synchronously.
That should eliminate caching of the filesystems in each UML, leaving the
host to cache stuff in its own memory.
> It appears that extra RAM improves things (in terms of how many uml
> instances you can run) - but not proportionally.
Inside UML or on the host? There a UML limit (and I forget what it is) which
I intend to get rid of.
From: William McEwan <mcewanw@ho...> - 2001-08-31 14:11:19
I'll check the below stuff out more once I'm back at work after the weekend
(the main computer is there).
> > It appears that extra RAM improves things (in terms of how many uml
> > instances you can run) - but not proportionally.
>Inside UML or on the host? There a UML limit (and I forget what it is)
>which I intend to get rid of.
I was referring to the memory in the host system - Ofer Wald's said his
machine was 256M RAM and he managed to get 20 umls running simultaneously -
my one had 384M and 31 umls ran okay. I'll try taking some memory out of
the host to check this effect.
Originally I thought that the uml memory (e.g. mem=32M) might be the decider
- I thought that by dividing the host virtual memory (i.e. RAM + Swap) by
the uml memory it might give a rough idea of how many umls would run - but a
quick check suggested that that didn't give the answer - I bumped up the uml
memory from 32M to 16M and, as far as I recall, I still couldn't run more
than the 31 umls. I didn't do any special tricks - except I commented out
the serial port getty in inittab - that, for some reason, seemed to allow a
few more umls to run without error messages (that is why I wondered if it
was anything to do with virtual console limits - though I don't know if that
question makes sense...). It could, as Ofer suggested be related to the
number of processes running on the host - though personally I think it is
some kind of memory paging problem that is causing the limit. I'll record
the error messages next time!
I did use COWd root filesystems.
How do you get the UMLs to mount their root filesystems synchronously?
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp