Cluster question

[omarrodriguezt]

Hello,

We tried to setup a cluster following poundtime guide http://www.poundteam.com/downloads/Vici ... v1%201.pdf but we were not able to have more than 20 agents running good.


I have one question:
When I set up the cluster the last time, two years ago, was:


1 - Database Server/ Webserver =>Dell PowerEdge
1 -Dialer => Dell PowerEdge 1950
1-Another dialer => Dell PowerEdge 1950


Dell PowerEdge 1950=> 8GB Ram, Core 2 Quad ==> http://www.dell.com/downloads/emea/prod ... t_Quad.pdf

The clusters works, but we were not able to handle more than 25 agents. So we decide that using the servers by separated will give us better results, and yes, the results were better. 15 agents per server, 3 servers, means 39 agents running fine. So the maths for us were not working, 1 server 15 agents vs 3 server in a cluster less than 25 agents.

But later we decide to rent servers with poundteam, with a similar story. We started having a database /webserver and another one dialer
(Memory 4G RAM CPU Intel Core2Quad 2.3+ Ghz.)
Having 2 servers in cluster we are able to handle no more than 23 agents.
Then we rent 3 more serves with poundteam and each one separated can handle 15 and 18 agents. So the maths are nor working again. 1 =15 but 3 is not 45

Also, when I set up the cluster the first time with 1 database/webserver and 2 dialers half of the agents were login in dialer1 and the other half in dialer2.
( I mean, all the agents were using the same webserver/database, but in order to use both dialer 10 agents were logged in dialer1 and the other 10 in dialer2, in their softphones)

But is this the best scenario? Because I thought that all the agents must be using same dialer configuration in their softphones (I mean registered to only one asterisk box in the eyebeam), so all agents login to the same webserver, same dialer, and the vicidial-asterisk will balance the load.

So, when you have a cluster working with more than 1 dialer, do all the agents log in just in one dialer, or you put half of the agents with the softphone registered to dialer1 and the other half registered the softphone in dialer2? If not, what I'm missing.

I'm telling you the story because we are planning to do the cluster again in order t have at least 60 agents working, but using
http://download.vicidial.com/iso/vicibo ... luster.pdf

Any comments based on your experiences will be appreciated.

[omarrodriguezt]

Is there anybody that can help us?
Your comments are appreciated it.

[omarrodriguezt]

So far, answering miself viewtopic.php?f=8&t=25952&p=90444&hilit=cluster#p90550

[spacejanitor]

Hi Omar,

You'll want to be a little more specific as to how each of the dialer servers are "unable to handle" more than those 20 agents. What were the symptoms you experienced? Your telephony servers are definitely very powerful, the PowerEdge 1950s are much beefier than my telephony servers... and my telephony servers comfortably handle 30-40 agents each with full call recording/transcoding to MP3.

So please tell us what symptoms you experienced in both situations, and what leads you to believe that the bottleneck is the telephony servers?

sj

[omarrodriguezt]

Today we have two power servers running one as database (pt37 (66)) and the other as dialer (pt38 (70)). This is the actual cluster, but this cluster can not handle more than 25 agents. We also have another 3 single stand alone servers (PTVTNL121 (93) , pt115(67) , pt74 (69)) that can handle 20 agents each one. So 2 servers on the cluster only 20~25, 2 single servers 40 agents.
Please help.
The server specs will be posted in a few minutes:

These are the cluster:
Data base Server pt37: http://pastebin.com/Bt3AFwRC
Server pt38 (70) - Dialer--> http://pastebin.com/kHqfUAA3

These are the specs for stand alone server:
Server pt115( 67.) http://pastebin.com/mFdVZJVy

[spacejanitor]

Today we have two power servers running one as database (pt37 (66)) and the other as dialer (pt38 (70)). This is the actual cluster, but this cluster can not handle more than 25 agents. We also have another 3 single stand alone servers (PTVTNL121 (93) , pt115(67) , pt74 (69)) that can handle 20 agents each one. So 2 servers on the cluster only 20~25, 2 single servers 40 agents.
Please help.
The server specs will be posted in a few minutes:
Data base Server pt37: http://pastebin.com/Bt3AFwRC

Server

You have 2 servers running a single database? How does that work, is the second one just a failover DB?

And you are running the Web Server together with the DB on the same machine? This is a little unorthodox. How is your system load on each of these machines when you are experiencing these issues (type 'htop' into shell).

Also you still haven't described the symptoms - please share these with us as well.

[omarrodriguezt]

You have 2 servers running a single database? No, 1 is DB and the other is Dialer.
How does that work, is the second one just a failover DB? no. 1 is DB and the other is Dialer? all the specs were posted.
And you are running the Web Server together with the DB on the same machine? yes.

Wow is your system load on each of these machines when you are experiencing these issues (type 'htop' into shell). We will post the result in a few seconds.

Also you still haven't described the symptoms - please share these with us as well.
We will share in a few minutes.
Did you take a look of the servers specs?

[spacejanitor]

Yes, saw that. Could use a little more RAM than 4GB, but shouldn't be a bottleneck at this time.

What kind of HDD(s) are you using and in what configuration?

[omarrodriguezt]

Disk /dev/sda: 500.1 GB, 500107862016 bytes
255 heads, 63 sectors/track, 60801 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Disk identifier: 0x0009c6ee

Device Boot Start End Blocks Id System
/dev/sda1 1 262 2104483+ 82 Linux swap / Solaris
/dev/sda2 * 263 60801 486279517+ 83 Linux

[spacejanitor]

Again I'm not seeing in there the type (SAS/SATA?) or configuration (RAID1/5/10... no RAID?).

[omarrodriguezt]

7200RPM SATA. not raid

[williamconley]

When you say "can only handle XX agents", what exactly do you mean by this? What symptoms do you experience? and have you attempted to qualify these symptoms? It could be that your DB server requires more connections, or any one of several "multi-server, high-volume" issues. It can be recommended that your HD be 15k rpm SAS (or at least a 6G/sec SATA) for the DB drive. There are also several DB configuration items to allow high-output for the DB server.

Best to find out where your bottleneck is and attack that instead of the "brake shop" approach of "let's fix everything that costs money and see if that helps". LOL

So: Attack your symptoms. One at a time.

And realize also that you are running on a Fairly Old version of Vicidial. Although also a Stable version (which is good) it is very old at this point. You're missing out on a lot of Toys.

Vicibox 4.0 will install a cluster from beginning to end, fully automated. The only part I don't like is the ntp setup, which is easily overridden to sync to a single local ibursted server. Consider a VMware cluster setup for practice and check out the my.cnf settings Kumba uses in the DB server of the cluster ... some of those settings may come in handy. But before all that: It would be best to actually know what you're chasing. If you are running out of memory on the Web server of the cluster, no amount of changes on the DB server will help you ... right?

[omarrodriguezt]

Thank you. That's a good idea.