Subscriber across the WAN

[tarun.pahuja]

Folks,

Cisco Doc says that to have a Subscriber across the WAN to a Publisher or a Cluster comprising of a Sub and a Pub we need atleast 768 Kbps. My question is as follows:

1) Do we need this bandwidth (approx 768 Kpbs) only for SQL replication or voice and control traffic can be included into it.

If i have a T1 circuit between 2 sites and the same senario, how much bandwidth would i reserve for SQL replication, real time replication, Voice and Control traffic.

Any Documentation on this would be highly appreciated.

Also, if one of the remote sites that needs to have a Subscriber happens to be across the globe (say japan), would someone recommend building a new cluster their (Pub) and connecting to the cluster using the gatekeeper or a Sub would be able to successfully replicate over a Frame connection between the two conuntries.

Thanks

[carl.newman]

Tarun:

From all of the reading and testing that I have done every thing points me to 900kbps of bandwith for a sub across a wan link. That same wan link must also be 40 ms or less. That would be really hard to get if the sub was in Japan and the Pub was in US. I would suggest that you look at a Cluster to Cluster solution with a gatekeeper. Newer version of CCM allow for a trunk to more than one member in a cluster. I think you would me much more stable to use a cluster to cluster model.

Carl

[tarun.pahuja]

thanks for the info.

Does that mean, i need to reserve 900K on my WAN link only for replication or it could be shared with voice and control traffic.

If i am using MQC to set up QOS, do i need to create a seprate calls for SQL replication and put 900k in priority queue. what is i only have a T1 connection?

if i remember correctly, there are 2 typed of replications that happen between the pub and the sub, does 900K account for both?

Any feedback would be highly appreciated.

Thanks

[jschweizer]

It is my experice that the you should use MQC to mark all traffic from sub to pub and vice versa to ip precedence 2 or DSCP AF21. This would be 768, or maybe a little more for safty. Then use MQC to with a service pol of 768 and three more classes one for Voice, one for control, and one called class-default(this one is bulit in just name it in the Service policy). The SQL doen't need high priorty but it sould be above all other nonvoice related traffic (i.e. call and control).

I recconend useing G.729 for the codec on the phone and this requires about 20k per call and you would need to figure out how many calls will cross the T-1.

[tpetzold_2]

What you need is at least 900 Kbps of QoS provisioned bandwidth for every 10,000 busy hour call attempts on the cluster. The connection between any two servers in the cluster must also be less then 20 milliseconds one way or 40 milliseconds round trip.

This bandwidth is not for SQL replication but for ICCS traffic (Intra-Cluster Communication Signaling). As someone else posted here you should provision extra bandwidth for SQL replication but that's not officially required.

All this is documented in detail in the CallManager 4.0 SRND (Solution Reference Network Design) guide posted here.

http://www.cisco.com/en/US/products/sw/voicesw/ps556/products_implementation_design_guide_book09186a00802c370c.html