[Fwd: HTT/SMP servers instability on 5.3-STABLE]

[RW]

On Friday 10 December 2004 19:57, Colin J. Raven wrote:
> On Dec 10, Toomas Aas launched this into the bitstream:
> > Colin J. Raven wrote:
> >> On Dec 10, Toomas Aas launched this into the bitstream:
> >>> For more than a month now, I've been running a Dual Xeon 2000 system
> >>> (IBM eServer xSeries 225) with 5.3-RC2 and later upgraded to
> >>> 5.3-RELEASE and I have no problems at all. Hyperthreading is disabled
> >>> in BIOS, ACPI is enabled.
> >>
> >> Excuse me for jumping into this thread but could you elaborate as to why
> >> you have Nyperthreading disabled?
> >
> > I'm afraid I can't give you any good technical reasons. I simply think of
> > HTT as Intel marketing blurb, meant to make you feel like you are getting
> > two CPUs for the price of one. Well, actually it's still only one CPU.
> > I've been running another single Xeon 2.4 box for more than a year.
> > Initially, I ran several months with HTT enabled. Then I ran several
> > months with HTT disabled. I didn't really notice any performance
> > difference.
>
> Well thanks for sharing those observations. Until I read what you said I
> *assumed* that there was a performance difference with HTT
> enabled. The specs seem to show that there *is* a theoretical
> difference, yet clearly according to your observations there just isn't
> any difference of earth shaking proportions.

Isn't the point of HTT that a CPU holds several threads from the *same* 
process. So if your software isn't multithreaded, there's no benifit. 

UNIX software tends to fork single-threaded process, rather than create new 
threads, so it benefits from SMP, but not HTT.