[c-nsp] WCCPv2 - what happens to existing connections when redirect-list is modified?

[Adrian Chadd]

On Fri, May 22, 2009, Dale Shaw wrote:

> Can anyone provide any insight?

> Adrian Chadd, I'm shining the bat torch towards the sky, are you out there? :-)

Sigh. Yes i'm here. :)

Unless stuff has changed, WCCPv2 will just still be matching on bits in your
packet headers and rewriting next hops.

The TCP state management stuff and redirection management stuff is done on
the cache engines (the WAAS boxes) rather than the routers themselves.

So if you update the redirect list, packets will most likely start flowing
to WAAS devices that are set to receive them (save say, any kind of
load balancing, slow start, etc which may be fiddling with your hash/mask
assignments in a way that rejects some packets) and then hopefully
the WAAS devices will do what the older school cache engines did:

* if its for that box, it'll terminate it locally;
* if its for another cache engine in the group that it -knew- about
  (and has seen the topology change happen), it'll redirect the packet
  to the cache that was last handling that flow;
* otherwise, it'll just punt it back to the router to be passed
  through.

But this is all a guess, I've not got a WAAS device to test and I've never
deployed them. :) The important bit to enlightenment here is exactly what
the router and cache engine responsibilities are. WCCPv2 pushes a lot of
the smarts (ie, everything session oriented) out to the cache engines,
leaving the router to get on with the job of punting packets.

2c,



Adrian