Choppy calls only for web logged in agents

[muyousif]

Hi, having one weird issue. All manual calls are fine but as soon as agents logged in via web and take calls, there is too much voice distortion. Latency to carrier is normal under 15 ms. Server load is fine, RAM is free, any idea what could be reason?




VERSION: 2.12-493a
BUILD: 150703-2105
© 2015 ViciDial Group

One DB, one Web, 4 Telephony servers with no extra card. Default vicibox installation. Total agents 200

[williamconley]

Manual dial calls often take a different path. Dial prefix difference?

[muyousif]

No dialplan is quite same. There is only one carrier through which all calls are going. When I listen to live calls agent voice is fine but customer voice is really low or choppy. But if I dial customer number through soft phone manually, voice is ok both way. any other idea?

[williamconley]

You did not post your installation method. This would be a good time to rectify that oversight. It may be relevant.

You also did not mention the hardware involved and you said certain items are "fine", but "fine" is subjective. It would be much better to post the actual values. For instance: Server load is defined by the 1,5,10 minute average server load available from "uptime/htop/top" and probably several other sources.

Is this server virtual? What's the hardware that it's on? (Pentium II for instance may not work very well with a scratch install ... LOL).

Next up: the dialplans are NOT the same between Manually dialed calls and autodialled calls. Even if there is one carrier configured, there may still be differences.

Most importantly: A meetme room is involved in one, but not the other. You may want to test your server's capability to hold a conference using a meetme room without the Login (dial one of the conference numbers directly ... and route an inbound call to it OR just call the same meetme room from two different agent phones).

And codecs ... are you using g729 or ulaw? Or something "else"?

[ambiorixg12]

These are my suggestions :

Use top command to monitor the CPU and memory usage while calling through the agent GUI and directly from the softphone.

Check codec used during the calls asterisk -x "sip show channels"

Check bandwith usage iftop -Nnp

[muyousif]

Thanks for the suggestions. I am continuously monitoring server load using htop and it is fine all the time never goes above 50%. Codec is Ulaw and there is no issue of bandwidth as we have fiber connection directly to ISP and download/upload is above 50 mbps.

1. Installation is done using vicibox 6.0.3 ISO so default installation no customization manually.
2. I have telephony servers (HP DL360 G5, 380 G5 and G6) having 4 core, 8 core and 16 cores processors. And according to the cores, Load is always below 50%. Like if 4 core then 1,5,10 minute average server load is below 2 and similar for other servers.
3. All servers are physical no virtualization.
4. Manual calling Dialplan

exten => _03.,1,AGI(agi-NVA_recording.agi,BOTH------Y---N---Y---N)
exten => _03.,n,Dial(sip/voipswitch/${EXTEN},,tTo)
exten => _03.,n,Hangup

Auto dialing dialplan

exten => _903XXXXXXXXX,1,AGI(agi://127.0.0.1:4577/call_log)
exten => _903XXXXXXXXX,2,Dial(sip/voipswitch/${EXTEN:1},,tTor)
exten => _903XXXXXXXXX,3,Hangup

So just a prefix (9) which is removed in the second line.

5. I am more concerned with meet me conference, We tried by dialing meet me conference from agents phones, 4 agents in same conference room and it went fine for about 10 minutes. I am going to try Amfeltec PCIe x1 card on the upcoming monday, hopefully it will work fine, if it doesn't then I have no option except going for smaller telephony servers as recommended by Kumba in a topic.

Thanks a lot for helping us to understand this awesome solution.

[muyousif]

Also i mean calling to meet me conference was fine all the time not only for 10 minutes. lol.

[williamconley]

A little more depth ... so when these calls are choppy, how many calls are live on the same server? Note that "Average" server load can be misleading if you have spikes in load that do not greatly affect the Average.

You mentioned three different servers with different core counts. Are they all dialers and all experience the same poor call quality regardless of the server load and/or number of other live calls?

If you have a multi-server installation, you could simply need to dedicate one (or more) server to "dialer only" and move DB/Web to other server(s) to avoid interference with calls. This does not require uninstalling anything, just stop using Web on the Dialers. If you have configured Apache for heavy load on that server, changing it back to normal would be reasonable.

I strongly suspect that your "manual dial" calls were tested when the servers were not otherwise in use, and you decided the servers were "ready", then you put Lots of calls on (with autodial) and got choppy calls ... please correct me if I'm wrong. Perhaps you should test ONE autodial called to properly mirror your ONE manual dial call scenario?

[muyousif]

Manual calls are tested during the period when agents were facing distortion on live calls. My database and web server both are separate from telephony servers. I will try to test servers separately. I have installed Amfeltec voice sync device in servers but dahdi_test results are not much satisfactory on all servers. I am not sure if DAHDI still using Dahdi_dummy or taking voice sync as timing device. Can you please give me some guideline how I can make sure voice sync device working?

[muyousif]

Got the config for Amfeltec.... need to change dahdi_dummy in vicidial initialization script /etc/init.d/vicidial...