Thanks Leo, great explanation. Will do some additional research and
try out a few tests if I can find the time to setup a small load-test
sort of a scenario but it does sound from your explanation that
symmetric multi-processing is what we need to share the load and get
double or close to double performance. Scheduling by its very nature
wouldn't be multi-tasking but rather a way to use up idle times to
perform more tasks and I don't believe there would be an idle
time/wait-time between a cpu-intensive task like transcoding in which
it'll get the time to use the "other" logical cpu to run another
transcoding operation. And the only thing I can think of why people
might be suggesting to turn HT off is because it has overhead and why
put up with it if we're not using it? Is that the thought behind it?
Anyway, thanks again. Looks like No VM and No HT for me :-{
\R
On 9/8/06, Leo Ann Boon <[EMAIL PROTECTED]> wrote:
Most virtualization platforms don't guarantee accurate timing. It's a
fundamental implementation issue. For example, if your host OS is
capable of 1000Hz. You won't expect the client OS to be able to do
1000Hz due to the overheads. Best case might be 999.9Hz, worse case
could be anything below that. IIRC, Xen VM used to do only 100Hz.
Asterisk is multi-threaded so SMP (either multi-way or multi-core) would
definitely help. But, HT would not. HT is still essentially still 1 CPU.
The CPU appears as 2 CPU by taking advantage of wait times to
multi-task. Good for normal usage like word processing, but bad when
running CPU-intensive tasks like codec transcoding.
_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --
asterisk-users mailing list
To UNSUBSCRIBE or update options visit:
http://lists.digium.com/mailman/listinfo/asterisk-users
