Op3r wrote:then 1 same spec for asterisk for every 25 agents (with recording and using ztdummy)
Its kinda overacting on the specs but at least it never gave me head aches. and that thing works perfectly
as for mysql clustering I never heard anyone who clustered the mysql database of vicidial.
Thanks for the info. I forgot to mention I'll use a blade (or two) for the web server, as you say. Also, right now, this is mostly an inbound call center so there is not much load from the predictive dialer. That will change in the future. Everything is SIP with G.729.
I want to cluster the MySQL just because I want it to never go down. We'd like to grow this call center to several hundred simultaneous agents (or more) so I want it to be as reliable as possible.
It seems to me that with a large number of agents, lets say 500, there is a lot of strain on the database and web servers due to the 1-update-per-second-per-agent-issue. 500 web hits and database updates per second is a lot. Hence I want redundancy and load sharing.
I'm really having a hard time with 25 agents per box. We're talking about a box with effectively 4 3GHz cpu's (HT off) lots of memory, and lots of FSB bandwidth. Surely 25 agents per CPU would be more realistic? But I'm asking this here because I haven't run a large vicidial installation.
If I really could do 100 agents per box, I calculate the recording bandwidth at 2.34MB/sec. Is that unrealistic?
My second thought is to buy an entire blade system on ebay, maybe the older 1855's, with 10 servers, older 3.0 GHz dual proc systems. At least the price/agent ratio would be better then.
Does eaccellerator help?
All opinions appreciated!
Mark