Cluster Install validation required

[pargat]

Hi,

My current setup has 300 agents, They are distributed across 3 servers: each server is stand alone single server vicidial install, each handling about 100 agents and having 4 E1 lines each. We see calls in queue and agents waiting during peak load.
This was handed over to me by my predecessor. And am still wondering why he did not cluster the setup.

I want to move this to a cluster ASAP as the setup is wrong, on a historic version and prone to breakdowns.
My basic requirements for the cluster are:

Support upto 400 Agents (Peakload: will happen once a month, normally will have 350 odd agents). The Process is inbound process, the outbound calls made by an agent are for 3 party conference. The inbound to outbound ratio is 3:1 (that is out of every 4 calls, 3 are inbound and 1 is outbound manual dial). There is no need of Predictive/ratio dialing out.
We need to use E1 lines for telco. Also all calls need to be recorded.

My plan is to have 2 active web servers, 1 active DB server, 1 backup DB server and 5 Asterisk servers each with 4 E1 lines. (Each Asterisk server will need to have a Sangoma/Digium card for E1 support)

I have the required servers with xeon quad core processors with 12GB RAM in each.

My questions are:
1) Is the above proposal for server distribution in the cluster good? If no what is recommended.
2) If ok, what should be recommended RAM in each of the servers mainly web and DB?

thanks a lot in advance.

[Kumba]

1x Database : 8 real CPU cores or more 2.1Ghz or faster, 32GB+ of RAM, Two SSD's in RAID1 under /srv/mysql for database
8x Telephony/Web : Quad-Core CPU, 8+GB of RAM, Digium cards for T1/E1 (7x Telephony, 1x Web)


I wouldn't advise going over 50 agents per server, but you can always add telephony servers later if you find them to be unstable. So you could start with 4 telephony servers and go from there.

[pargat]

Big thanks for your response Kumba!!

My reasons of going for 5 Asterisk servers were:
1) No predictive/ratio dailing, hopper does not come into picture, all inbound calls and if required an outbound 3 way conference.
2) E1 lines to be used, so lot of voice processing and handling is on the digium/sangoma card.
3) Existing setup (vicidial release from 2011) has 4 E1 lines per server (approx 120 channels) and 75 to 90 agents per server working quite well.
4) Our average talk time on an inbound call( we have 100% inbound and only manual outbound for 3 way conference) is 2minutes plus.

Also what I plan to do is have dual homed machines for the cluster. All machines in the cluster talk to each other via a dedicated switch, while as all agent communication works over another switch. Hopefully this will keep the inter server communication faster.

Thanks once again for your inputs, much appreciated.