Re: Top not showing 4 cpus on 2 xeons with HT

Adrian Chadd wrote:
anyone have any recent information about this? some people say "HT
sucks for almost all workloads", others say "recent scheduler
improvements make HT more useful".. is there anything reasonably
authoritative?

No, because it depends on your applications and workload.
for some it is better, for some it is not. Therefore it's
best you try both variants on your own machine with your
own applications and measure the difference.

There's one rule of thumb, however: If you only have one
HTT-capable processor, then a UP kernel will almost always
be the better option, because the locking overhead of an
SMP kernel will probably outweigh any advantages of HTT.
On the other hand, if you have a real SMP system (i.e.
multiple processors or cores, not counting HTT), then you
will want to use an SMP kernel anyway. In that case,
enabling HTT will probably not hurt -- _but_ there have
been reports of some people that HTT hurts in such a case
for certain kinds of applications (I think databases was
one of them, but I don't remember exactly).

Anyway, there are exceptions to any rule, so you should
measure yourself.

Personally I disable HTT on all of my machines because of
the security issue (jails do _not_ help here at all!),
and speed improvements -- if any -- are marginally small,
according to my own measurements. In fact I had a hard
time finding any reproducible measurable improvements at
all for my typical workloads; consequently my decision
was governed by the security issue.

Best regards
Oliver

--
Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing b. M.
Handelsregister: Registergericht Muenchen, HRA 74606, Geschäftsfuehrung:
secnetix Verwaltungsgesellsch. mbH, Handelsregister: Registergericht Mün-
chen, HRB 125758, Geschäftsführer: Maik Bachmann, Olaf Erb, Ralf Gebhart

FreeBSD-Dienstleistungen, -Produkte und mehr: http://www.secnetix.de/bsd

"Perl will consistently give you what you want,
unless what you want is consistency."
-- Larry Wall
_______________________________________________
freebsd-stable@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-stable
To unsubscribe, send any mail to "freebsd-stable-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: Hyperthreading in 6.x ... still frowned upon?
    ... If you've seen "HT" or "HTT" applied to an AMD system, it's likely an abbreviation for "HyperTransport" or "HyperTransport Technology". ... The workloads I've seen the best improvement in performance for HTT have been ones involving a healthy blend of floating point and integer instruction mixes, or ones with a lot of memory stalls. ... My recommendation would be to evaluate the performance impact of HTT against your specific workload and see what impact it has, but not be surprised if it doesn't help. ...
    (freebsd-stable)
  • Re: HTT and SMP question
    ... > that did the trick. ... Note that HTT causes worse performance in many workloads, ...
    (freebsd-stable)
  • Re: HTT and SMP question
    ... >> that did the trick. ... >> regards, ... > Note that HTT causes worse performance in many workloads, ... Would you please point out some cases when HTT with machdep.hlt_logical_cpus=0 ...
    (freebsd-stable)
  • Re: HTT and SMP question
    ... >> Note that HTT causes worse performance in many workloads, ... >> that sysctl defaults to off. ... Kris ...
    (freebsd-stable)