[Asterisk-Users] Comprehensive Asterisk Load Testing

[Kristian Kielhofner]

Hello,

I would like to test the capabilities of the various hardware that I
run AstLinux on:

- Soekris Net4801 (266mhz Geode)
- 1ghz P3
- 1ghz Via C3
- 2.5ghz Celeron
- 3 ghz x 2 Xeon


What I would like to do is use * on the higher end machines to pound as
many calls as possible (probably 10, 20 at a time) into * on the lesser
machines. I will then try to keep track of system resources (CPU usage,
memory usage, etc) on the "client" machines. I want to do this with
various codecs, jitterbuffer yes/no, trunk yes/no, SIP, IAX, across all
of this hardware to at least get an idea of what I can expect from these
CPU's (as far as transcoding goes). "show translations" is just not
cutting it anymore... :)

Not to self-promote, but AstLinux looks like a perfect platform to do
testing like this because of consistency and the fact that it can run
from flash and RAM, so disk I/O should not ever be a problem...

I am thinking some combination of app_milliwatt & the outgoing call
spool or manager interface would be a good way to go about this. The
wiki page has no specifics for doing this, so I thought I would ask.
How is this normally done, or is there a completely different, better
way to do it?


Thanks in advance!

--
Kristian Kielhofner

[Zoa]

Hey,

Most of the time you dont need a big machine to test a small machine.
Just make sure there is no transcoding on the sending end.
I did all the tests you mentioned (Except for the jitter buffer) on a
dual xeon and a via c3.

That took me about 2 months fulltime (its a lot harder than it looks),
you can find some of the results on www.astertest.com (there you will
find also some imature version of a callgenerator for asterisk that
would probably help you to do things faster).
I could also help you off list if you want.

Zoa.

Post by Kristian Kielhofner
Hello,
I would like to test the capabilities of the various hardware that
- Soekris Net4801 (266mhz Geode)
- 1ghz P3
- 1ghz Via C3
- 2.5ghz Celeron
- 3 ghz x 2 Xeon
What I would like to do is use * on the higher end machines to
pound as many calls as possible (probably 10, 20 at a time) into * on
the lesser machines. I will then try to keep track of system
resources (CPU usage, memory usage, etc) on the "client" machines. I
want to do this with various codecs, jitterbuffer yes/no, trunk
yes/no, SIP, IAX, across all of this hardware to at least get an idea
of what I can expect from these CPU's (as far as transcoding goes).
"show translations" is just not cutting it anymore... :)
Not to self-promote, but AstLinux looks like a perfect platform to
do testing like this because of consistency and the fact that it can
run from flash and RAM, so disk I/O should not ever be a problem...
I am thinking some combination of app_milliwatt & the outgoing
call spool or manager interface would be a good way to go about this.
The wiki page has no specifics for doing this, so I thought I would
ask. How is this normally done, or is there a completely different,
better way to do it?
Thanks in advance!
--
Kristian Kielhofner
_______________________________________________
Asterisk-Users mailing list
http://lists.digium.com/mailman/listinfo/asterisk-users
http://lists.digium.com/mailman/listinfo/asterisk-users

[Bicom Systems]

Post by Zoa
Hey,
Most of the time you dont need a big machine to test a small machine.
Just make sure there is no transcoding on the sending end.
I did all the tests you mentioned (Except for the jitter buffer) on a
dual xeon and a via c3.
That took me about 2 months fulltime (its a lot harder than it looks),
you can find some of the results on www.astertest.com (there you will
find also some imature version of a callgenerator for asterisk that
would probably help you to do things faster).
I could also help you off list if you want.

Zoa..

Have you done the test using call generator on test or production boxes?

Ta
Senad

[Zoa]

Its a very very bad idea to do this on production boxes. Especially if
you are trying to see how far you can go, and then you cross that tiny
border :)

Your production calls will not like an idle cpu% of 0% and a load of 500.

zoa,

Post by Bicom Systems

Post by Zoa
Hey,
Most of the time you dont need a big machine to test a small machine.
Just make sure there is no transcoding on the sending end.
I did all the tests you mentioned (Except for the jitter buffer) on a
dual xeon and a via c3.
That took me about 2 months fulltime (its a lot harder than it looks),
you can find some of the results on www.astertest.com (there you will
find also some imature version of a callgenerator for asterisk that
would probably help you to do things faster).
I could also help you off list if you want.

Zoa..
Have you done the test using call generator on test or production boxes?
Ta
Senad
_______________________________________________
Asterisk-Users mailing list
http://lists.digium.com/mailman/listinfo/asterisk-users
http://lists.digium.com/mailman/listinfo/asterisk-users

[Bicom Systems]

Post by Zoa
Its a very very bad idea to do this on production boxes. Especially if
you are trying to see how far you can go, and then you cross that tiny
border :)
Your production calls will not like an idle cpu% of 0% and a load of
500.

I could not agree more with you hence my question :)

However, the tests results produced on test boxes:
How realistic it is?
Does it really presents "real life"
scenarios and results?
Does it take in consideration different
type of services (calls, IVR, queues) ?

I am not trying to put down anyone or anything here, I am just
curious.

Ta
Senad

[Kristian Kielhofner]

Post by Bicom Systems

Post by Zoa
Its a very very bad idea to do this on production boxes. Especially if
you are trying to see how far you can go, and then you cross that tiny
border :)
Your production calls will not like an idle cpu% of 0% and a load of
500.

I could not agree more with you hence my question :)
How realistic it is?
Does it really presents "real life"
scenarios and results?
Does it take in consideration different
type of services (calls, IVR, queues) ?
I am not trying to put down anyone or anything here, I am just
curious.
Ta
Senad

Senad,

I have yet to take a real hard look or contact Zoa, but if all you are
doing is calling an extension (very rapidly and many, many times) it
really would not be very hard to test queues, music on hold, meetme,
etc. I am downloading the callgenerator from astertest.com right now...

The most realistic test is to (obviously) register as many phones as
possible and hire hundreds of people to talk on them... :)

--
Kristian Kielhofner

[Matt]

Has anyone gotten this tester to work? i can get it to log in and
show me my call load.. but it doesn't seem to MAKE any calls.

Post by Kristian Kielhofner

Post by Bicom Systems

Post by Zoa
Its a very very bad idea to do this on production boxes. Especially if
you are trying to see how far you can go, and then you cross that tiny
border :)
Your production calls will not like an idle cpu% of 0% and a load of
500.

I could not agree more with you hence my question :)
How realistic it is?
Does it really presents "real life"
scenarios and results?
Does it take in consideration different
type of services (calls, IVR, queues) ?
I am not trying to put down anyone or anything here, I am just
curious.
Ta
Senad

Senad,
I have yet to take a real hard look or contact Zoa, but if all you are
doing is calling an extension (very rapidly and many, many times) it
really would not be very hard to test queues, music on hold, meetme,
etc. I am downloading the callgenerator from astertest.com right now...
The most realistic test is to (obviously) register as many phones as
possible and hire hundreds of people to talk on them... :)
--
Kristian Kielhofner
_______________________________________________
Asterisk-Users mailing list
http://lists.digium.com/mailman/listinfo/asterisk-users
http://lists.digium.com/mailman/listinfo/asterisk-users

[Zoa]

The call generator is very not user friendly now and undocumented, i
recommend not to use it and use some simple script somewhere for now.
(you could find some scripts somewhere on astertest.com).
I will put fixing that callgenerator on the (big) todo list.

Zoa,

Post by Matt
Has anyone gotten this tester to work? i can get it to log in and
show me my call load.. but it doesn't seem to MAKE any calls.

Post by Kristian Kielhofner

Post by Bicom Systems

Post by Zoa
Its a very very bad idea to do this on production boxes. Especially if
you are trying to see how far you can go, and then you cross that tiny
border :)
Your production calls will not like an idle cpu% of 0% and a load of
500.

I could not agree more with you hence my question :)
How realistic it is?
Does it really presents "real life"
scenarios and results?
Does it take in consideration different
type of services (calls, IVR, queues) ?
I am not trying to put down anyone or anything here, I am just
curious.
Ta
Senad

Senad,
I have yet to take a real hard look or contact Zoa, but if all you are
doing is calling an extension (very rapidly and many, many times) it
really would not be very hard to test queues, music on hold, meetme,
etc. I am downloading the callgenerator from astertest.com right now...
The most realistic test is to (obviously) register as many phones as
possible and hire hundreds of people to talk on them... :)
--
Kristian Kielhofner
_______________________________________________
Asterisk-Users mailing list
http://lists.digium.com/mailman/listinfo/asterisk-users
http://lists.digium.com/mailman/listinfo/asterisk-users

_______________________________________________
Asterisk-Users mailing list
http://lists.digium.com/mailman/listinfo/asterisk-users
http://lists.digium.com/mailman/listinfo/asterisk-users