Page 1 of 1

Dual Processor

PostPosted: Fri Sep 08, 2006 3:01 am
by marvin
Does anybody here tried using vicidial on a dual processor environment?
How does it affect the whole server performance?
Did the number of seats increased compared to single proc?

PostPosted: Fri Sep 08, 2006 6:18 am
by Op3r
90% of echosystem (philippine based asterisk guys) install of vicidial are on dual xeon boxes.

The increase of performance? it really boils down to the proper configuration.

PostPosted: Fri Sep 08, 2006 6:44 am
by mflorell
The number of seats will increase, but not proportionally to having double processing power.

Performance also depends greatly on what kind of setting you have and what kind of VOIP codecs you are using if any.

PostPosted: Fri Sep 08, 2006 9:11 pm
by vctor
Yes we already have installed vicidial on a dual processor, unfortunately we did not get the performance boost that we were expecting on a dual proc system.

PostPosted: Fri Sep 08, 2006 9:42 pm
by marvin
ic ... thanks guys!

whats the reason behind this? anybody knows the reason or perhaps a theory behind this? ... just like vctor, i was expecting that if you use those dual proc your performance will be more or less equal to those dual server ... i really dont know much in hardware

PostPosted: Fri Sep 08, 2006 9:53 pm
by mflorell
two reasons:
- motherboard BUS speed and the limitations of the path from the CPU to the memory
- The load spikes from Asterisk seem to grow exponentially not proportionally when they happen.

PostPosted: Sat Sep 09, 2006 4:30 pm
by gschaller
The path from CPU to memory is bad on Intel processors. AMD processors are directly connected to memory, so they are faster. Intel dual-core processors have to share the slow path to memory. I have a Asterisk on a AMD 4400+ (dual-core 2,2 GHz with 1 MB cache/core). On my two server setup (Asterisk + Apache/MySQL) the Asterisk server has a loadavg of maximum 0.7 with 10 Agents, full recording and dial-level 2.0. I think I can have minimum 20 agents on this setup. At the beginning I had a dual-core Xeon (Dell Poweredge SC1420 with 3 GHz I think, don't buy this one, isn't compatible because the chipset has a timing problem). This server had a loadavg of 1.6 to 1.8. Anyone with a newer Intel processor?

PostPosted: Sat Sep 09, 2006 6:10 pm
by mflorell
I just ordered an Intel "Core 2 Duo" processor which is supposed to address a lot of the data path and bottleneck issues of the first generation of Intel Dual core processors. Won't be in for two weeks. I will post my results when I get it all running.

PostPosted: Sat Sep 09, 2006 11:36 pm
by gerski
Intel "Core 2 Duo" processor runs perfectly on asterisk, im now running 14 agents with no codec transcoding. and the loads is less than 1.0 :lol:

with 2gig Memory
Sata Hardisk
Realtek Lancard
Digium X100P

PostPosted: Sun Sep 10, 2006 12:16 pm
by marvin
i have here AMD Athlon(tm) 64 Processor 3000+ running with max of 15 agents before funny things happens :lol: no recording though.

just recieved the xeon 3.4 that i ordered couple of weeks ago ... ill try to hit on this sometime this week ... am expecting that this would reach at least 20 seats just like the FAQ says.

PostPosted: Sun Sep 10, 2006 5:59 pm
by kchung
Here's an idea: create a benchmarking script which simulates X amount of meetme conferences and Y amounts of loopback dialing to simulate predictive dialing and Z amounts of database queries done by the vicicdial client.

PostPosted: Sun Sep 10, 2006 7:30 pm
by mflorell
That's my plan. After 2.0.1 is released of course.

PostPosted: Sun Sep 10, 2006 8:35 pm
by kchung
Awesome!
and of course create a website where we can submit data as a community. Perhaps the script can automate the submission, along with hardware and software profile and a checksum to make sure no one is BS'ing?

PostPosted: Mon Sep 11, 2006 8:54 am
by mflorell
That's a bit more than I was planning to do. Benchmarking will differ a lot depending on the setup that different people use. Some have only Zap channels, some have on only SIP channels. Some people record all calls and some record none. It's very hard to simulate every kind of setup. I was just planning to test a few different setups all with the same hardware to show what the capacity and/or load for that kind of setup would be like.