i dont have the exact details, but I would the core would have to support numerous tcp connections, probably a 6371 to support 50k tcp optimized connections.

Or load balance across smaller appliances

I have worked larger deployments of WAAS, and all have similar head ends. One of my deployments have currently over 900 unites heading to over 2000.

All of my large deployments have multiple large boxes (WAE-7371s) that are load balanced by hardware platforms performing WCCP (6500/7600).

Most of these deployments also have connectivity directives from all the edges to 1 or 2 datacenters for file services as well as general optimization. These can be grouped into similar BW and Latency groups for management.

That give you some ideas?

Dan

How do the bandwidth and latency group approaches work? Are you referring to grouping the core clusters together like that? Are you using any sort of QOS on the CIFS 4050 traffic or other optimizations?

I don't usually use QOS outside what is normally on the WAN.

I group similar bandwidths and latencies to each DC that you want to optimize to. So you may have 20-30 T1s from 20-100ms. You can create a group called T1-1-14ms, a group called T1-15-30ms, T1-31-45ms, etc. I loosely group these on how many 4050 sockets they open up to the core. That way you can just add WAEs to the groups as you install and activate them. They just then inherite the configurations.

Thisshould go away with transparent CIFS in 4.1.

Dan

Do the latency settings across groups impact the overall performance of CIFS across those links when configurations are different under the connectivity sections? For instance, does mismatching the 80ms/1500Kbps default with a 160ms/1500Kbps WAN affect performance greatly?

They can, for instance in your scenario, WAAS will attempt to adjust the buffers to send more CIFS data across the 4050 connections before waiting for responses, using the concept of BDP (bandwidth delay product). The edge will also open more 4050 connections for edge to core communications.

Dan