Gigabit on UCM Server NICs?

Unanswered Question
Apr 20th, 2010

A customer is finally upgrading from 4.x to 7.x and they're wondering whether to spare Gig ports for the new servers, questioning whether the UC servers really need the bandwidth. The topology is a centralized, six-server solution (Pub, 3 subs, 2 TFTP) supporting ~3000 devices (and expanding rapidly), including approximately 40 MGCP gateways.

I couldn't find anything in the SRND, so my first question is: what are the bandwidth requirements and the recommended NIC speed to go with on the servers? Next, do they really need teamed GIG interfaces with load-balancing for the expected traffic?

The premium (all gig and teamed) requires 12 Gigabit ports off their core switches versus 6 Gig ports for single, or keeping them on 100Mb if there's really no reason to allocate the ports.

Assistance appreciated.



I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
htluo Tue, 04/20/2010 - 13:40

If you really want the details, you'll have to consider the following:

1) How many phones are registered to a specific CUCM (registration, keepalives, signaling).

2) Is the CUCM going to be a TFTP server (configuration download).

3) Is the CUCM going to be a MoH server (media streams)


I would say go for the GE, unless you're really running out of GE ports.  Having a GE connection doesn't mean CUCM is going to consume gigabit bandwidth.

Teaming is more like a redundancy other than bandwidth increase.


Brian Carscadden Wed, 04/21/2010 - 06:27

The customer is looking for something more concrete. They don't want to commit Gig ports if they don't have to. Considering the cluster size, as well as Unity servers, Contact Center, Presence, Meetingplace and Cisco Fax - the ports add up. Add teaming and we're looking at 26 ports of gigabit. Granted some of these adjunct servers can be bumped to 100Mb.

Is there a guide on bandwidth requirements available from Cisco?

I understand some possible metrics involved, including TFTP, MoH, the number of phones, intracluster communications, etc..

Any additional info would be appreciated.



Brian Carscadden Wed, 04/21/2010 - 07:45

Thanks. I checked this document and although it provides generally good advice, it doesn't specifically speak to

the bandwidth requirements or service considerations within the cluster in terms of traffic.

htluo Wed, 04/21/2010 - 08:04

"Bandwidth" is a debatable topic (ie. there's no right or wrong answer).

"Bandwidth" is more meaningful when the application has a consistent/continuing traffic (like video streaming).

For non-consistent/non-continuing traffic, queue management and TCP windowing play a more important role.

Sadly, Cisco design guide didn't give a detailed formula to calculate bandwidth requirement.  You may be able to find more info on other IPT books like "Cisco CallManager Fundamental", "CIPT QOS" etc.

Generally speaking, I've never seen a bottle neck on 100M interface in CUCM deployment.  To take precautions, you might want to separate media stream and signaling stream on different CUCM servers.  That should help preventing traffic jam.

Hope this helps!



This Discussion