experiencing some delayed audio playback on outbound survey

[callforamerica]

Hi,

I've been running into a problem of audio playback being delayed when caller answers phone, using routing extension 8366 without AMD..

I stripped it down to 75 max trunks per server and 50 calls per second.

I'm running 6x 2core 4gb ram servers on Asterisk 11.22.0-vici &

VERSION: 2.12-565a
BUILD: 160827-0917

It doesn't appear to be a MYSQL problem. I've tuned mysql. checked kernel recycling of UDP ports.

Are there any recommendations on what to look for?

How many max trunks per server should it be able to handle.

These are all press-1 campaigns just broadcasting messages. There are no agents answering calls or being transfered. It's being managed by remote agents.

Performance overall on server looks fine when problem is occuring.. under 13% cpu, plenty of free ram.

[mflorell]

How many servers do you have?

Do you have each server set to 50 CPS(calls per second)?

What has your carrier said your maximum CPS to them is?

How many leads are in your vicidial_list table?

What are the hardware specs on your DB server for this cluster?

[callforamerica]

6 servers

Yes, CPS set to 50 and Max Trunk 75. Carrier has 200 CPS & 2000 Trunks, I haven't had any failures and they report I haven't exceeded 100 CPS.

VIcidial_list is 700k

DB server is xeon 3.2ghz with 16gb ram

No 'conference' agents logged in ever.

Technically I would think each server could be maxed out to at least 100 trunks, no?

[mflorell]

50 CPS per server is WAY too high. There is no reason to have it set that high at all. You would probably be fine setting it to 5-10 CPS.

What are the hard drive and RAID card specs on your DB server?

Do you have slow query logging enabled on your DB server?

Yes, you can probably get 100 max trunks on each dialer with those specs.

[callforamerica]

It's a raid 5

slow query is on, but no surprising entries.

It's pretty consistently doing it, even when i ramp each server down to 35 trunks.

DB server always stays under 15% and I keep monitoring on it.

It's almost like the perl process for AGI is delayed.

[mflorell]

RAID-5 = VERY BAD FOR VICIDIAL DATABASES!!!!

Also, you didn't mention what RAID card you are using.

You should be using a RAID-1 or RAID-10 for a VICIdial database.

[callforamerica]

It ended up being my database server. After the call is connected, it writes to vicidial_log, and this query was taking 8+ secs. Im still working to fix it.

Thanks for your help