Re: [uml-user] swap clarification

[Howard G. <Ho...@at...>]

More questions that may help point toward an answer.

1) Is memory allocated to a UML locked out from the host?
   ex. If you run UML with mem=512M option and the UML is just running an 
ftp server how much memory will the UML allocate

2) If you have several UML's on a host would it be preferable to prevent 
one UML from taking too much memory?
   ex. Host has 1GB and you have four UML's
      -- If each UML is run with mem=512M it seems that two of the machines 
could hog all of the memory from the host and starve the other two machines 
and the host as well.
      -- If you instead configured each UML with mem=200M and a 300M swap 
file, it seems each of the machines would always be guaranteed it's 200M 
and the host would always be left with at least 200M.

3)  Is it a good idea to tell a machine it has "fast" memory (aka. RAM) 
when in fact it has "slow" memory (aka. Disk)?
   Does the following make sense or am I missing something?
      -- Linux seems to like to use all available "extra" memory for disk 
cache so I would suspect that over time, a UML would end up trying to use 
all of the RAM it thinks it has as disk cache.  If this "RAM Cache" is 
in-fact swapped out to the host's hard drive, then it seems like you would 
be hurting your disk performance in the UML.  When it read a sector from 
the ubd device and attempted to cache it it would trigger a disk write on 
the host.
     -- I think I am assuming that a kernel (UML or otherwise) is smart 
enough to not try to use swap memory to cache hard drive information but a 
UML would not be able to know if using what it had been told was RAM would 
actually cause a memory swap to disk.

-Howard

At 10:28 AM 3/8/2002 +0100, Adrian Phillips wrote:
> >>>>> "Patrick" == Patrick Thomas <us...@ut...> writes:
>
>     Patrick> I have read the following on the web page:
>
>     Patrick> "If needed, and it has swap configured, it will swap"
>     Patrick> (from: kernel.html)
>     Patrick> ---
>
>     Patrick> my assumption has always been that when I run a UML
>     Patrick> instance on my computer (or two or three) that swapping,
>     Patrick> if it is needed, will be taken care of by the underlying
>     Patrick> machine ... this belief was fostered by the HOWTO which
>     Patrick> never instructs you to do anything swap related.
>
>     Patrick> So the question is, am I correct ?  Does my UML instance
>     Patrick> (and the other two running on the same machine) have
>     Patrick> swap, or do I have to explicitly define internal UML swap
>     Patrick> for each of them ?
>
>I was just thinking that myself !! I'm running bonnie++ now in the uml
>to see the difference between host disk access and uml access. The
>though of swapping/caching popped into my head as I was running top in
>the uml.
>
>It would seem to me that (as with vmware) you would give the uml as
>much memory as you think it would need but no swap and let the host
>linux handle the swapping issues. On the other hand, perhaps in some
>situations giving UML access to a raw device for swapping would be
>better ?
>
>  I don't know how this would work with caching though, presumably you
>would want to sync in the uml relatively often so that the host can
>sync the uml's file(s) otherwise you'll have files in the uml's
>filesystem taking up to twice as long to sync to disk. For some reason
>I though there was an option to do syncing in uml but I can't find it
>now or perhaps it just mentioned mounting with sync in the uml ?
>
>It would be good to get some "tuning" points in the HOWTO by the way;
>I'd be certainly willing to write something (in plain text at least)
>about swap if we can clarify what is deemed sensible.
>
>Sincerely,
>
>Adrian Phillips
>
>--
>Your mouse has moved.
>Windows NT must be restarted for the change to take effect.
>Reboot now?  [OK]
>
>_______________________________________________
>User-mode-linux-user mailing list
>Use...@li...
>https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user