CUCM Design and Deployment of CUCM Services (tftp, cti manager, moh)

[calmichael]

After countless reviews of SRNDs over time, I am just looking for something that goes to current best practices on CUCM cluster sizing and the distribution of CUCM services like tftp, cti manager, moh, etc.

Discussions with peers and Cisco SE/CSEs (presales and TAC) have yielded alot of different opinions on this.

e.g.

The presales process has procured 5 7845-H2 servers and licensing for 7000 IP Phone users for a centralized deployment for a distributed enterprise.

Other voice applications are VM (CUC) for the 7000 users and CC (UCCX) for 175 users.

There was no high level design outside of a Cisco BOM to utilize unspent budget so assume:

The high level design implies 1:1 redundancy though not stated.

The high level design implies scalability but does not state any growth count or growth trend.

The high level design does not characterize call volume.

How should the CUCM cluster designed and the services (tftp, cti manager, moh, etc) deployed (activated) as a low level design is being developed?

Opinions researched and debated seem to fall into three groups:

1) Dedicated Publisher node (one physical) + Dedicated 1:1 Subscriber nodes (two physical) + Dedicated 1:1 tftp/cti/moh nodes (two physical)

2) Dedicated Publisher node (one physical) + Two Non-dedicated 1:1 Subscriber/tftp/cti/moh nodes (four physical) with CallManager Groups creating two logical 1:1 groups to spread ~3500 users on each logical group.

3) Dedicated Publisher node (one physical) + Three Non-dedicated N:1 Subscriber/tftp/cti/moh nodes (three physical) with CallManager Groups creating three logical N:1 groups to spread ~2333 users on each logical group; the tftp/cti/moh would be backup + One Dedicated tftp/cti/moh node (one physical) the tftp/cti/moh would be primary.

Opinions or links?

[William Bell]

Given the parameters I would think that the main sticking point is 7,000 stations. The 7845 is marketed as being able to support 7,500 stations. I don't typically subscribe to this model. Typically, I go with a 5,000 station threshold before I look at expanding the cluster. But that is my opinion. Putting that aside for a moment, I would look at:

(1) Publisher. Dedicated

(2) TFTP/MoH/Software Media servers. Dedicated. Meaning, the IP Voice Media Streaming Service and the TFTP service is enabled on these systems

(2) Subscriber/CTI Manager servers. Dedicated. You MUST run CallManager service on a CTI Manager host.

If you were being conservative (like I am) and you want to keep a 1:1 model, then you would have two additional servers for Subscriber/CTI Manager. If that is too expensive, then you could do a 2:1 model (still one extra server) and balance the load across both primary CUCM subscribers. Be aware of the different failover scenarios.

Interestingly enough, I haven't found a definitive recommendation in the SRND, but I have found a specific recommendation in the Release Notes for CUCM 7.1.3. I found this helpful. Hopefully it will help you as well:

http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/rel_notes/7_1_3/cucm-rel_notes-713.html#wp1791070

HTH.

Regards,

Bill

[James Hawkins]

Hi,

Is your query "How do I implement this with five 7845-H2 servers?" or do you have the possibility of ordering the extra one/two servers required for Bill's recommendation?

[David Hailey]

To touch on a different point - you mentioned additional services such as voicemail for 7000 users.

With Unity Connection 7x (on appropriate hardware/e.g. 7845), you can support up to 10,000 users in a voicemail only deployment. In an Integrated Messaging (IMAP) deployment, 7500 is the supported limit in terms of users. Personally, for that size deployment and given that you are interested in redundancy - I would go with a Cisco Unity Connection cluster which is a 1:1 active/active HA model where you have a Pub and a Sub. In normal operations, the Pub is the primary DB and Web servicing node and the Sub is the primary call processing node. If calls burst (say during a peak hour), calls can rollover to the Pub. If one server fails, the other assumes all operations until the failed node comes back online. All data is replicated between both nodes.

Hailey

Please rate helpful posts!

[calmichael]

This is a fictional scenario only; but the fundamental questions about cluster roles and service deployment still remain.

This debate started at a client site which Cisco Advanced Services originally built, was then supported by the partner channeland was critiqued by a Cisco DE working with TAC in troubleshooting "performance issues" when a bug was suspected versus design.

Investigating the issue further, we could not find anything specific outside of defining the various services and a few "you should get a dedicated publisher at this amount of phones or agents" or "you should get a dedicated MOH at this number of streams" nuggets; but nothing failry complete since the Golden Bridge documents could be found online.

Partners tend to have their own knowledge capture mechanisms, but I was hoping to discover the missing link for something Cisco published either online or recorded at a seminar or event (e.g. Live!)

I almost felt I made the e.g. sound to close to a VUE question; sorry about that .

I chose 7000 users as I knew the limit to be 7500 on the 7845-H2s used in the e.g. left less than 10% scalability capacity.  I mentioned VM and CC as I know CC via UCCX causes some CTI consideration.  I mentioned the lack of call volume as BHCA was one of the typical inputs for this line of questioning. All these were to keep the question under some control and guidance.

Thanks for the reponses!