New 500 seats installation

[floydans]

We are a planning a new 500 seats installation of vicidial based on vicibox.

only outbound campaign
dial ratio=3
available tally=Y
recording on demand (average 10% of active conversations)


This is what We are thinking of setting up:

1x DB Server (with slave replication for reporting purposes)
4x QuadCore Xeon CPUs
32 GB Ram
4 x SAS 146GB 15k rpm RAID 10
Keepalives = 12345678

10 x Vicidial/Asterisk Dialer Server (1 server each 50 agents)
1 x Xeon CPU QuadCore
4 GB Ram
2 x SAS 146GB 10k rpm RAID 1
Keepalives = X

2 x Apache Web Server
1 x Xeon CPU QuadCore
4 GB Ram
2 x SAS 146GB 10k rpm RAID 1


As for external lines we will use 25 x E1 (25x30 lines with a total of 750 lines).
We are still in doubt whether distribute those E1 across the 10 dialer servers or use trunks from dialer servers to three central asterisk servers with 8 port Sangoma cards.


Hope your experience will put me on the good road. Any advice on correct server size, distribution and architecture will be really appreciated.

[williamconley]

If you intend to install a 500 seat Vicidial based system, you NEED to call The Vicidial Group and give them a consultation fee to assist in the selection and configuration of your system. If not just plain have THEM configure and install all the servers.

If would be a serious lack of prudence to stick to "free advice" on a system of that magnitude when the makers of the software are available.

I suspect that the amount of $$ you are about to spend on this system would qualify as an "investment". Further, I suspect that the $$ invested in the training and staff and equipment aside from this system directly are also "on the line". Not having experts (when available) to assist in this endeavor is not the approach *I* would recommend, honestly.

No: I do not get paid from or by The Vicidial Group. But if *I* were to set up a 500 seat call center, *I* would call them to assist with design. Without hesitation. And not because I don't know what I'm doing, because I do , but simply because the investment is too large to risk not having the advice of the true experts. Not to mention taking advantage of ALL the resources within the software that are available.

[floydans]

That's what we actually did before posting...we did ask VicidialGroup of course.

Let me add that we already some experience as we do install and run 200 agents sites. But I was curious and interested in others experience. If there is anyone there running an installation over 200 seats It would be nice to share information as installation method, version, hardware manufacturer and model, DB server and Dialer Server sizing...etc...

[mflorell]

At a minimum I would recommend doubling the RAM in your DB server and adding one more Apache web server.

I might even suggest getting a few Acard RAM drives with 32GB RAM each for the DB so that you can have faster disk access to the database on the DB server:
http://www.acard.com/english/fb01-produ ... dno_no=270

[floydans]

Thanks Matt for replying. Actually I was having a look at acard drives (found on one of yur previous post:
http://www.vicidial.org/VICIDIALforum/v ... hp?t=15319).

Still have a question, would you use Acard Drives only for Mysql datadir or would you install also OS of the DB server?

Would I achieve the goal if I install OS on 2xSAS 15k rpm (Raid 1) and then datadir on Acard drives?

[mflorell]

It really depends on how much money you have to spend

I would recommend using them for the data directory only, but if you have the budget you might want to use them for the root filesystem as well.

If you do that it would be a good idea to have some regular drives for logs and backups on the server as well.

[Op3r]

Here's some tips also on getting VICIDIAL stable for 500 seats center. (*Note I am currently the IT Manager of a 600+ seats call center here in the Philippines.)

VICIDIAL cluster per campaign.

Some outbound campaigns only have 10 agents some 30 agents some 50 agents and the biggest i have right now is 280 agents.

What I usually do is asks the expansion plan. If it will do 40 agents per shift from the start, I usually create a new cluster for it. If it is only 20 agents I do single server.

You will also need at least a team of dialer people who will upload leads /modify campaigns/check system status and a whole lot more for VICIDIAL.

When it comes to hardware, it all boils down to a quad core for the asterisks/vicidial and the biggest and the baddest server you can get your hand's on for the database. I use an IBM m2 series with 320gb hard drive on raid10, 32gb ram and centos 64bit.

Although it's a lot of servers to manage and purchase, the cost of VICIDIAL being fully functional and head-ache free can seriously help bring the CAPEX and OPEX lower than what you should be spending if you are using a closed source system.

And yes I usually get drunk on friday nights, can sleep far more than 8 hours a day and have a hobby other than computers.

(this is the longest post I have ever made since when I became a member of the forum.)

[williamconley]

How do you handle agents moving between campaigns? (or do they just not *do* that?)

[Op3r]

Usually the practice is when the account get's canned the agents are let go. The agents are trained on a particular products/accounts exclusively.

The agents going to another campaign entails another training session to get familiarize with the stuff they are doing outbound for, this gives us allowance to create additional servers to handle the agents.

Sometimes multiple campaigns are hosted on a cluster hence the planning stage for the account. If the agents are required to log in to different campaigns on any given day on the same account(client) we host it on a cluster.


and for the phones, we have floor plans for particular campaigns.

[DarknessBBB]

At a minimum I would recommend doubling the RAM in your DB server and adding one more Apache web server.

I might even suggest getting a few Acard RAM drives with 32GB RAM each for the DB so that you can have faster disk access to the database on the DB server:
http://www.acard.com/english/fb01-produ ... dno_no=270

Hi Matt, in a specialized storage usenet group they don't talk very well about this hardware.

# Data transfer rate up to 400MB/sec
# IOPS 20,000 per SATA port

Is not so much

They suggest, for the same price, to use this:
http://www.ocztechnology.com/ocz-revodr ... s-ssd.html

240GB-960GB Max Performance

Read: Up to 740 MB/s
Write: Up to 720 MB/s
Sustained Write: Up to 600 MB/s
Random Write 4KB (Aligned): 120,000 IOPS

They also suggesto to use this engine for SSD drives:
http://www.rethinkdb.com/

What do you think?

[mflorell]

Data transfer rate is relatively unimportant, the important part is access speed and write speed for ViciDial MySQL usage, this is why RAM is ideal, and much faster than SSD.

[floydans]

Thanks Op3r for sharing your experience.
I'm not sure I got what you mean when you say "cluster per campaign".
Do you mean you have a DB Server wit for CPUs and keep adding and removing Dialer Servers from the cluster depending on your needs...?

Or maybe you mean you have a single DB server with multiple mysql instances for multiple vicidial clusters?

[williamconley]

he likely has a sep mysql server for each cluster, since that's the least expensive

it is also possible to have a sep DB for each cluster, but all on the same server ... but that doesn't make sense unless we are keeping the data segregated for a multi-tenant situation (which he does not seem to be in).

but multiple sql "instances" implies virtual mysql, and for this size installation, that would be a bad idea.

[floydans]

So this means Op3r manage a 600+ seats CallCenter across different vicidial clusters, each with its own db server+dialer server... the bigger being 280 agents...?

Some outbound campaigns only have 10 agents some 30 agents some 50 agents and the biggest i have right now is 280 agents.

[williamconley]

sounds like it to me. 280 is a fairly large cluster. trying to get bigger than that in one cluster is a great way to lose hair.

[Op3r]

Yep that is correct.

For every campaign we create a cluster for them.

This also minimizes the downtime.

So for example campaign A is down it wont affect the whole freaking floor.

[floydans]

I use an IBM m2 series with 320gb hard drive on raid10, 32gb ram and centos 64bit

Hey Op3r, how many CPUs do you have onto this server ??

And...Do you use this server for the larger cluster only and use smaller DB severs for smaller clusters?

[DarknessBBB]

I use an IBM m2 series with 320gb hard drive on raid10, 32gb ram and centos 64bit

Hey Op3r, how many CPUs do you have onto this server ??

And...Do you use this server for the larger cluster only and use smaller DB severs for smaller clusters?

And... do you use VoIP based phone lines, right?

[Op3r]

the dbase IBM have 16 cores.

and yes I use voip.

If ever I will need to use TDM/PSTN Im making a switch.