Re: signifanctly slowdown of FreeBSD 8.0-CURRENT/amd64

Kip Macy wrote:
I'm running the r193133 amd64 with a custom kernel and all debugging off
on an AMD Athlon64 3400+ single-core, and I haven't noticed any significant
slowing, although I haven't been doing any systematic benchmarking.

What would be the penalties of running an SMP -CURRENT kernel on
single-core hardware with no hyperthreading? Can anyone quantify the
typical added overhead? Or, counterintuitively, would an SMP kernel
be better in some ways?



He is trying to diagnose if the problem was introduced by enabling
adaptive spinning on sx locks. They're only enabled on SMP kernels.

Cheers,
Kip

Well, no SMP on UP AMD CPUs without SMT means no usage of the sx locks.
As far as I know, the sx locks were introduced a couple of days ago,
weren't they?

To avoid any kind of misunderstanding, there is no permanent 'slowdown',
it occurs especially whenever the system's compiler builds world or a
kernel or heavy I/O activities occur. It seems my boxes, especially the
UP box, is freezing, no reaction on X11. Well, first time I thought it
is related to UP, low memory or especially new drm code used with X11
for acceleration, but I also realized those 'slowdown-incidents' on
boxes with multicores, 8 or more GB of RAM and no X11 installed and
running. This performance impact in situations whenever building a
world is significant. We did not change the compiler, it is still the
old gcc 4.2, so I do not expect an impact from new high-memory and
cpu-consuming optimization routines. I do not even benchmark my boxes
day for day, but I usually do a set of the same work on those boxes used
for scientific work, so while building world even on SMP boxes and
working after installation with the same software set reveals some
weirdness in some points.

I thought this is due some use-uninterruptable debugging switches, some
wrote the malloc code is about to change and so on so I was wondering if
there is something temporarely going on at the moment where some hints
could prevent me from building a world within this testing phase. Just
speculation.

Greetings,
Oliver
_______________________________________________
freebsd-current@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: Goodbye ASUS, Hello Gigabyte
    ... unless an smp kernel, smp doscall1.dll, and smp os2loadr are used. ... you have to use an SMP kernel and the SMP DOSCALL1.DLL. ... I thought this was actually good, in my case, because I couldn't use a longer DVD R/W optical drive with the ASUS board in my case because the end of the drive hit the memory modules on the ASUS board while on this one they didn't. ... a long video card, it will block the IDE connector. ...
    (comp.os.os2.setup.misc)
  • Re: Cluster computing drawbacks
    ... |> SMP CPUs by spawning multiple processes and threads. ... boxes of a cluster. ... a few CPUs, for reasons that we could go into but are irrelevant. ... The point is that, if processes don't communicate, they can perfectly ...
    (comp.arch)
  • Re: Latest FC kernel still have SMP bugs?
    ... > Dual Xeon servers with HT on supermicro server boards. ... loop-back mount/umount until you get sick of it and ^C out. ... Any FC1 SMP kernel I have used on several types of systems, ... AMD single processor systems running the athlon SMP kernel, ...
    (Fedora)
  • [PATCH] x86: merge asm-x86/alternative.h
    ... * Alternative inline assembly for SMP. ... * works fine for patching a SMP kernel for UP. ... +static inline void alternatives_smp_switch ... * Alternative instructions for different CPU types or capabilities. ...
    (Linux-Kernel)
  • Please confirm: top(1) cpu time broken on FreeBSD 6.1 (RELEASE/STABLE) SMP ?
    ... upgraded SMP 2 CPU boxes. ... Boxes are upgraded from 5.5 to ... All of them have SMP kernels, just using the provided GENERIC and SMP ...
    (freebsd-stable)