I have two design queries relating to wccp service groups and WAAS in DC and branch deployments.
Firstly, lets say at the DC end I use wccp service 61 (source address) on the WAN interface of my edge-layer switches. I configure the L3 interfaces on the same switches (connecting to the LAN side) to use use wccp service 62 (destination address). The WAEs are using L2 at the edge layer; with masking etc.
I've read that at the branch office you need to 'reverse' the service group setup - so that if I have the same sort of setup at the branch using 3750s and WAEs then the WAN interface should be using wccp service 62 and the LAN side using wccp service 61.
If I assume that is correct, then how does this affect things when two branches are communicating with each other (and they are both setup the same) - will be waas not be effective in this scenario? (Assume that the DC waas does not see any branch-to-branch communication).
What happens if you have a consistent design across your network (61 on WAN, and 62 at LAN interfaces across all WAAS sites)?
Secondly, when using L2 wccp redirection and masking; do most deployments leave the mask as default (0x1741)? I'm thinking that in some situations it might be better to have an entire geographic location covering a few branches being sent to the same DC end WAE. For example, I might want everyone on a /24 subnet in one branch to be using the same WAE/dre cache at the DC; rather than the possibility of duplicate dre caches on DC end WAEs service the same branch subnet (I realise that redundancy might be an advantage should one DC WAE fail).
Is there a table/calculator somewhere that can work out what mask I could use to cover /24 or /22 or even /16 subnets to direct requests to the same WAE at the DC?
Excellent questions. Rule of thumb is to use source IP based load balancing, so in the branches 61/LAN - 62/WAN and in the DC 61/WAN and 62/LAN. That being said, if there is some site to site traffic at the edges, you may get some splitting, however, unless there is enough traffic to make it a "mini-dc", changing the services around is generally a wash. Also, if you only have a single WAE at the edge, it won't matter either.
On the mask, default mask is definitely not desirable. I generally use Calc and convert my desired Mask from Binary to Hex. The following examples are assuming 4 bit masks, but you can use from 1 up to 6 or 7 max bits if you need more buckets.
If you are looking to group /24, you could be 0xF00 or similar.
If you are looking to group /22, use 0x3C00 or similar
When calculating your mask, don't put your bits in the host bits, only in the network bits. Also, remember that the leftmost bit is usually the decision maker, so don't make it too far to the left or all your traffic will be on one WAE. The less WAEs in your WCCP cluster, the less bits you should use in your mask (allow some extras for fault tolerence).
Hope that helps,