Load issue on vicibox cluster

[javad-afzal]

Hi
we had a vicibox express install. we were having load issues due to high record database (Lists include around 20 Million records for now). so we shifted our database to a new server. Now the cluster includes Web + Asterisk on one machine and DB on the other one. Both are connected at the same gigabit switch over same local network.
We are still facing load issues on the Web + Asterisk Machine when call volume is around 50 to 60 Inbound calls with the CIDLOOKUP method enabled on campaigns and ingroup (It is only Inbound campaign with 50 Agent logins with CIDLOOKUP enabled and campaign recording is also turned on for the calls). All logged-in agents eventually log out from the agent portal when there is a high call volume on it. Calls still remain connected but the agent portal needs to be logged in again. Our cluster settings are mentioned below. Using Ulaw codec on asterisk.

192.168.2.X (asterisk + Web)
ViciBox_v9.x86_64-9.0.2.ISO | Vicidial 2.14-751a Build 200425-0949 | Asterisk 13.29.2-vici | Web + Asterisk Server | No Digium/Sangoma Hardware | No Extra Software After Installation | Intel(R) Xeon(R) CPU X5675 | 32 GB RAM |
VARactive_keepalives = 123456789E

192.168.2.X DB
ViciBox_v9.x86_64-9.0.2.ISO | Vicidial 2.14-751a Build 200425-0949 | Asterisk 13.29.2-vici | DB Server | No Digium/Sangoma Hardware | No Extra Software After Installation | Intel(R) Xeon(R) CPU E5-2670 v2 | 128 GB RAM | RAID 1 over SAS SSD | Load average below 1%

VARactive_keepalives = X

Also made below changes on both of the servers in my.cnf
max_connections=2000
Key_buffer_size=16GB

I am unable to understand where is the issue now. sometimes load average on the asterisk server is normal even with 30 connected calls, and sometimes it goes above 5% even with 15 to 20 inbound calls. Is there any issue with the DB settings or should i also move the web to the DB server.

Getting below warning on the database server frequently.
[Warning] Aborted connection 2871087 to db: 'asterisk' user: 'cron' host: '192.168.2.X (Dialer servers IP address)' (Got an error reading communication packets)

[carpenox]

My suggestion would either be to separate the web from the telephony with that many agents on it or update asterisk to 13.34.0-vici....or both

[javad-afzal]

I have separated the web from the asterisk server and using it through the DB server now. The setup now contains only an asterisk on 1 server and Web + Db on the other one. Now the load is fine on both of the servers i.e around 1% average. but still, we are facing the same issue, the agent goes in DISPO status (On realtime campaign summary page) there connected calls do not disconnect when it happens but they have to re-login again to take the next calls as the agent web portal stops working for them.
Gone through the apache2 logs as well, nothing seems irregular there as well. Are the below-mentioned settings for my.cnf good? for a database with around 10 to 15M records.

max_connections=2000
Key_buffer_size=16GB
Kindly do let me know where to look further as i am stuck for now.
Thank you

[carpenox]

that sounds like a dialplan issue, what does your carrier setup look like? if the call is not disconnecting, it would make sense the "hangup" isnt working right on your system