[c-nsp] Riverbed and WAAS in the path - will they clobber each other?

[Dale Shaw]

Hi all,

I have a curly scenario that some of you may be able to help with.

I have a situation where I need to add Cisco WAAS WAN optimisers to a
set of WAN links that already have Riverbed appliances installed
in-line.

I hope this turns out:

Data Centre:
 .--------.
 | Server |
 `--------'
     |
.-------.  .------.
| C6500 |--| WAAS |
`-------'  `------'
   | .----------.  .---------.
   `-| Riverbed |--| 7206VXR |--> WAN
     `----------'  `---------'

Branch Office:
                      .------.
                      | WAAS |
                      `------'
                         |
        .---------.  .-------.
 WAN <--| 2811ISR |--| C3750 |
        `---------'  `-------'
                         |
                    .--------.
                    | Client |
                    `--------'

I will be using WCCP to send traffic to the WAAS box at both the head
end and customer sites. Traffic flow will be as expected:

Client --> Server:
Client -> Cat3750 -> redirect -> WAAS -> return -> Cat3750 -> 2811ISR -> WAN
WAN -> 7206VXR -> (pass through Riverbed) -> C6500 -> redirect -> WAAS
-> return -> C6500 -> Server

Server --> Client:
Server -> C6500 -> redirect -> WAAS -> return -> C6500 -> (pass
through Riverbed) -> 7206VXR -> WAN
WAN -> 2811ISR -> Cat3750 -> redirect -> WAAS -> return -> Cat3750 -> Client

My query relates to the fact that the Riverbed will 'see'
WAAS-optimised TCP traffic flying through it. I'm not intimately
familiar with how each vendor mangles TCP, but I know Cisco use TCP
options and I assume Riverbed do too.

Obviously there are some branch offices with Riverbed and some (soon
to be) with WAAS. At the head-end, I can ensure that only traffic to
known WAAS sites is intercepted, meaning that the in-line Riverbed can
just "do its thing" as it currently does. But do I need to explicitly
tell Riverbed to ignore traffic destined for known WAAS sites? Is it
likely to 'clobber' or otherwise mangle WAAS traffic?

I would hope that both vendors' implementations will play nicely with
each other, but you never know until you try. By sending this message
I hope to become enlightened and, perhaps, avoid some pain and
suffering.

cheers,
Dale