[Asterisk-biz] Forklift a 2000 phone PBX - 5000 calls up - WAY up

[Kevin P. Fleming]

alex at pilosoft.com wrote:

> I'm still confused. Is Paul handling (media and all, not just SIP
> referring) 5000 calls on Asterisk? If so, on *any* kind of hardware,
> that's impressive. If not, it's not impressive at all, on any kind of
> hardware.

It's not clear to me at all; some of us "Digium geeks" discussed this at 
VON and a few times since then, and I have not yet seen any detailed 
report of the test methodology. Paul's response from earlier in this 
thread implies they did have media transfer, but people have told me 
that SIPP does not do media at all, and their published 'report' says 
they used SIPP.

I would love to see the test methodology and actual measured numbers 
reported, both so we can learn just how good their system may be, and 
also so others can run the equivalent test on different platforms to 
learn just what is important in handling call volume. For example, I did 
not mention the clock speed difference between the two platforms they 
tested, because everyone knows that CPU clock speed is one the less 
important factors here, behind memory bandwidth, I/O bandwidth and 
interrupt response latency (also, their Itanium 2 box has only a 400MHz 
FSB, but since it is 64 bits wide I figure that's comparably to a P4's 
800MHz FSB).

Paul's comments about the 6.4GB/s bus also intrigue me: my understanding 
of the Altix platform is that this high-speed bus is used for 
node-to-node interconnect, not internally within the nodes. If that is 
true, then testing on a single 2-CPU 'brick' doesn't bring the 
high-speed bus into the picture at all. The same would be true if 
someone ran a test on a single-node Cray XD1; sure they have an insanely 
fast interconnect bus (Myrinet, or something similar), but it's not used 
unless you are running multiple nodes.