Re: RHEL4 running on VMWare ESX 2.5.2

[Tom Sightler <ttsig tuxyturvy com>]

On Wed, 2005-10-26 at 13:05 -0400, Eric Hagberg MorganStanley com wrote:
> On Wed, 26 Oct 2005, Tom Sightler wrote:
> 
> The noht flag doesn't do anything for RHEL4 kernels.
> 
> See /usr/src/linux/arch/i386/kernel/setup.c on an RHEL3 vs. an RHEL4
> machine to see that it's gone.
> 
> And in that same file, there's an interesting note (under RHEL4):
> 
>                 /*
>                  * If the BIOS enumerates physical processors before logical,
>                  * maxcpus=N at enumeration-time can be used to disable HT.
>                  */
>                 else if (!memcmp(from, "maxcpus=", 8)) {
>                         extern unsigned int maxcpus;
> 
> Bummer if your bios doesn't enumerate physical before logical procs, eh?
> Or if you thought that "noht" flag actually disables hyperthreading under
> RHEL4.

You are correct.  Did this change during the beta, or am I'm just
remembering this from another linux distribution.  Sure enough, in my
notes on RHEL4 I wrote down that this parameter seems to have no effect
I apologize for bringing it up.  I specifically remember noting that the
"noht" option was useful on one distro because otherwise Linux mapped a
shared runqueue over the two virtual processors rather than two
different CPU's.

It's possible that this is not even an issue for RHEL4.  Even though it
appears that RHEL4 detects both CPU's as a single processor with a
physical ID of 0 and a sibling count of 2, which seems wrong, our
testing seems to indicate that the scheduler balances tasks fairly well
anyway.  Maybe I'm misinterpreting what this information means.

I'm all ears about this as we are currently testing some RHEL4 servers
on ESX 2.5.2 and hoping to get our hands on a pre-release of ESX 3 so
that we can take RHEL4 for a spin there as well.

Later,
Tom