Re: [LUG] 8GB of RAM in 32-bit machine...

[tom <tompotts@xxxxxxxxxxxxxxxxxxxxxx>]

On 26/06/10 08:43, Gordon Henderson wrote:
> On Fri, 25 Jun 2010, Max Siegieda wrote:
>
> > On Fri, Jun 25, 2010 at 10:27 AM, Gordon Henderson
> > <gordon+dcglug@xxxxxxxxxx<gordon%2Bdcglug@xxxxxxxxxx>
> > > wrote:
> >
> > > I may have a requirement to stuff an existing server with more than 
> > > it's
> > > current 4GB of RAM - it's running 32-bit Debian Lenny. The mobo will 
> > > cope
> > > and it's a new Intel core-2 processor.
> > >
> > > Now, I'm not after a single process with more than the current 3GB 
> > > limit,
> > > but I may need to run several programms which individually want a 
> > > few GB -
> > > hence putting more memory in it - basically to stop it swapping.
> > >
> > > (It's a "server" for a specific customer application, so not very 
> > > general
> > > purpose, but it is running "LAMP", but no X, etc.)
> > >
> > > However, I've no experience of more than 4GB of RAM in a 32-bit 
> > > system, so
> > > just wondering if anyones done this - basically what I'm trying to 
> > > do is
> > > avoid a complete re-install on the box (which is 300 miles away) 
> > > with 64-bit
> > > Debian (although compiling a kernel to cope with the added memory is 
> > > fine)
> > >
> > > Any experiences? good/bad/indifferent?
>
> > I'm no expert but every report and my own experience has been that 
> > you cant
> > have 4GB of RAM in a 32bit machine, physically it'll fit but it wont 
> > be in
> > use and it's a case of luck as to how much beyond 2GB is registered. The
> > limit is not 3GB per process.
>
> I suspect you're thinking of Windows... AIUI You need the server 
> editions of XP, etc. to be able to see more han 4GB of RAM in a 32-bit 
> machine. Linux sees it fine (or so I'm told)
>
> As for memory in a single process, well, a simple loop of malloc 1K at 
> a time gives me:
>
>   3044 Got 3117336 MB
>
> So about 3GB.
>
> Output from 'top' on that box:
>
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
> 25248 root      20   0 3069m 3.0g  348 S    0 87.3   0:01.85 a.out
>
> and if I run a 2nd, it gets the same:
>
>   3045 Got 3118090 MB
>
> (actually it got 1KB more!)
>
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
> 25378 root      20   0 3070m 2.9g  296 S    0 85.8   0:04.00 a.out
> 25248 root      20   0 3069m 152m  260 S    0  4.3   0:01.85 a.out
>
> but note that nearly all the first's been swapped out :)
>
> (the program ends by looping on sleep(100); )
>
> I think there's some architecture issue with 3GB being the max. 
> process size in a 32-bot machine (rather then up to 4GB as you might 
> expect), but I'm not that clued-up on the details.
>
> Gordon
I think the 3GB per process is a 'kernel' thing 1GB reserved for the 
kernel process' ?  IF you have a Physical Address Extension ( a new 
core2 processor should have that but can your motherboard utilise it?) 
capable processor you can use more that 4 GB of ram on a 32 bit machine 
but the most a single process can have is 3GB
You can recompile the kernel to use less than the default 1GB and get 
about 3.7GB per process which might help. I would question what using 
that much though as filling ram is a good way of locking things up (for 
a period of time)
Tom te tom te tom

--
The Mailing List for the Devon & Cornwall LUG
http://mailman.dclug.org.uk/listinfo/list
FAQ: http://www.dcglug.org.uk/linux_adm/list-faq.html