Hi all,
I'm using 'show translation' to help dimension my system, but I
confused
by the results I get. My 2 test systems (results below): an AthlonXP
2000+ (1.3GHz) and a Pentium D930 (duo-core, 3.0GHz) produced similar
results (D930 is slightly faster). Googling shows that others have
similar results running on other CPU speeds >2.0GHz.
At first glance, it would look like the AthlonXP gives better bang for
the buck :). But, I'm sure that are other reasons. I know show
translation times how long it takes a convert 1s of full duplex audio. I
suspect the test is using a single CPU (since it's in a single thread)
and there are some constant overheads that makes a 3.0GHz produce the
same numbers as a 1.3GHz.
I would love to hear how others are using the results from show
translation in system dimensioning. So far, I feel that dimensioning an
Asterisk box is still mostly guesstimation :). Currently, I'm using the
30MHz per call rule to dimension.
Leo
Results from show translation:
On the athlon:
g723 gsm ulaw alaw g726 adpcm slin lpc10 g729 speex ilbc
g723 - 3 2 2 2 2 1 3 10 11 13
gsm - - 2 2 2 2 1 3 10 11 13
ulaw - 3 - 1 2 2 1 3 10 11 13
alaw - 3 1 - 2 2 1 3 10 11 13
g726 - 3 2 2 - 2 1 3 10 11 13
adpcm - 3 2 2 2 - 1 3 10 11 13
slin - 2 1 1 1 1 - 2 9 10 12
lpc10 - 4 3 3 3 3 2 - 11 12 14
g729 - 4 3 3 3 3 2 4 - 12 14
speex - 4 3 3 3 3 2 4 11 - 14
ilbc - 4 3 3 3 3 2 4 11 12 -
On the D930:
g723 gsm ulaw alaw g726 adpcm slin lpc10 g729 speex ilbc
g723 - - - - - - - - - - -
gsm - - 2 2 2 2 1 5 9 10 14
ulaw - 2 - 1 2 2 1 5 9 10 14
alaw - 2 1 - 2 2 1 5 9 10 14
g726 - 2 2 2 - 2 1 5 9 10 14
adpcm - 2 2 2 2 - 1 5 9 10 14
slin - 1 1 1 1 1 - 4 8 9 13
lpc10 - 3 3 3 3 3 2 - 10 11 15
g729 - 3 3 3 3 3 2 6 - 11 15
speex - 3 3 3 3 3 2 6 10 - 15
ilbc - 3 3 3 3 3 2 6 10 11 -
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.digium.com/pipermail/asterisk-users/attachments/20061222/63b0287e/attachment.htm
Eric "ManxPower" Wieling
2006-Dec-22 21:34 UTC
[asterisk-users] How accurate is show translation?
Leo Ann Boon wrote:> Hi all, > > I'm using 'show translation' to help dimension my system, but I confused > by the results I get. My 2 test systems (results below): an AthlonXP > 2000+ (1.3GHz) and a Pentium D930 (duo-core, 3.0GHz) produced similar > results (D930 is slightly faster). Googling shows that others have > similar results running on other CPU speeds >2.0GHz.Does "show translation recalc 30" show any different results?
On 12/23/06 09:51 Leo Ann Boon said the following:> I would love to hear how others are using the results from show > translation in system dimensioning. So far, I feel that dimensioning an > Asterisk box is still mostly guesstimation :). Currently, I'm using the > 30MHz per call rule to dimension.on a Pentium D 2.80Ghz, we've sustained 300 simultaneous IAX2 calls terminating in a dialplan loop that answers the call, waits 2 seconds, plays demo-instruct and loops again. a cursory examination revealed that a large portion of the CPU was used to handle NIC interrupts. occasionally we got a chan_iax2.so error which said, "Maximum trunk data space exceeded to..." this seems to be controlled by the MAX_TRUNKDATA constant in chan_iax2.c which is set to 40ms of SLIN for 200 calls. it'd be nice to know what this constant is for and what would the implications of increasing it be. [cc'ed to -dev as well] -- Regards, /\_/\ "All dogs go to heaven." dinesh@alphaque.com (0 0) http://www.openmalaysiablog.com/ +==========================----oOO--(_)--OOo----==========================+ | for a in past present future; do | | for b in clients employers associates relatives neighbours pets; do | | echo "The opinions here in no way reflect the opinions of my $a $b." | | done; done | +=========================================================================+
On Sat, Dec 23, 2006 at 09:51:24AM +0800, Leo Ann Boon wrote:> Hi all, > > I'm using 'show translation' to help dimension my system, but I confused > by the results I get. My 2 test systems (results below): an AthlonXP > 2000+ (1.3GHz) and a Pentium D930 (duo-core, 3.0GHz) produced similar > results (D930 is slightly faster). Googling shows that others have > similar results running on other CPU speeds >2.0GHz. > > At first glance, it would look like the AthlonXP gives better bang for > the buck :). But, I'm sure that are other reasons. I know show > translation times how long it takes a convert 1s of full duplex audio. I > suspect the test is using a single CPU (since it's in a single thread) > and there are some constant overheads that makes a 3.0GHz produce the > same numbers as a 1.3GHz.If you had just one call, then adding extra CPUs wouldn't have helped. 'show translations' mainly helps you compare different codecs. It is also handy as a benchmark because it's there. However> > I would love to hear how others are using the results from show > translation in system dimensioning. So far, I feel that dimensioning an > Asterisk box is still mostly guesstimation :). Currently, I'm using the > 30MHz per call rule to dimension.There are some other factos. For instance, if you test relatively short calls (as someone else in this thread did), then the call set-up and tear-down overheads carry a larger wheight. It is also significant if you have an expensive dialplan (e.g: running an AGI for every call). -- Tzafrir Cohen icq#16849755 jabber:tzafrir@jabber.org +972-50-7952406 mailto:tzafrir.cohen@xorcom.com http://www.xorcom.com iax:guest@local.xorcom.com/tzafrir
On Sun, 24 Dec 2006, James Harper wrote:>> When you built Asterisk, it must have refused to build the ilbc codec > - >> I have never seen an Asterisk box that could not transcode ilbc, in > over >> 3 years of working with Asterisk. >> > > Ah... I use debian, and they tend to have pretty strict policies on > anything that isn't free (as in speech). Without having looked into it > further, that would probably explain it.I use Debian too (sarge), and I got ilibc for nothing ... A quick check on the sources shows it's included with Asterisk, so shouldn't depend on anything Debian has (or hasn't) got, other than a C compiler... Maybe you are using Debians packaged version? I don't, but prefer to compile my own. Gordon