Database Scalation / dimensioning

[covarrubiasgg]

Guys, sorry to post this since it is already a F.A.Q. but i can not find specific answer for my question. Also sorry because this will be a long post

The Vicibox Installation manual says this:

Recommended Database specification :
Approximately 150 agents, or 300 agents with double the amount of RAM and CPU cores•Eight core CPU 2.0+ Ghz•16+ GB RAM (More RAM generally gives better DB performance)•500+ GB RAID1 drive configuration for the OS•250+ GB SSD RAID10 drive configuration mounted under /srv/mysql for MySQL

Can I Scale this up to 1,000 agents ?

I mean can I assume this?

150 Agents
Eight core CPU 2.0+ Ghz•16+ GB RAM

300 Agents
16 core CPU 2.0+ Ghz•
32+ GB RAM

450 Agents
24 core CPU 2.0+ Ghz•
64+ GB RAM

600 Agents
32 core CPU 2.0+ Ghz•
128+ GB RAM

750 Agents
40 core CPU 2.0+ Ghz•
256+ GB RAM

1000 Agents
48 core CPU 2.0+ Ghz•
512+ GB RAM

So following that specs, my current setup will only handle 300 Agents

2x Intel® Xeon® Processor E5-2620 v4 ( 8 real cores each - 16 cores in total for the Dual CPU Setup )
128 ECC RAM
LSI 3108 MegaRAID w RAID 1+0 8 x 256 Gb SSD

I build this for less of $4,000 USD ( The whole server including case, disks, controller, etc etc )

If i want to get up to 600 Agents, i need to double the cores, so I would require Dual Xeon® E5-2683 v4 which costs about $2,000 USD each CPU, so for the costs of the CPU I can build a second server for 300 agents.

If I want to get up to 1,000 things start to get ridiculous. That would require Dual Xeon® E5-2699 v4 which cost $4,200(ish) each, so $8,400 only on CPUs, also at this point I will need a more expensive Motherboard and power supply in order to handle the wattage and heat. This build easily can be $15,000 USD price tag, 4 x times the cost of the 300 agent build.

Is it worth to spent that much on Single BD Server ?
Is there any strategy to use 4 x Servers with specs 300 Agents each ?
Did I misunderstood the dimensioning of the database ?

Please note that this will be running with daily archive and nightly optimize tables;

[mflorell]

We generally recommend separate clusters of up to 300 agents with transfers done using Cross-Cluster-Communication so that you can transfer calls with lead information between the clusters if you need to.

That said, our largest single cluster on a single master DB server was 600 agents, and that was pushing it, and you have to keep tight control on your table sizes(like vicidial_list) at that level.

What exactly will these 1,000 agents be doing on this system, will they all be on one campaign, with the same roles?

[covarrubiasgg]

I asked this as a general question to get a better understanding on best practices to dimension a cluster beyond 300 agents which is my biggest cluster so far. So you have already provided a very good starting point to me, i have never used the Cross-Cluster-Communication you mentioned.

But the project I'm working now is already working with 300 Agents, one single campaign but multiple Inbound Groups. This is a 90% Inbound / 10% Manual Outbound, they have already rent space for another 300 agents by the end of the year and looking for the 1,000 agents on before 2023

I think this can be easily splitted into different clusters with the only downside of having to run reports on different clusters, but for the sake of performance I think it worth it. Maybe I will have to create a custom report that queries the 3 clusters simultaneously.

[mflorell]

Doubling the CPU cores and RAM of the 300-agent DB specs should work for the 600 agent cluster. As for 1,000 agents, we've never done a single cluster that large so I can't say definitively that doubling that would work or not.

As for your 90/10 inbound/manual-dial cluster:
What is the anticipated daily call volume?
How many max calls waiting in queue will this system have?