WAAS Farm design

jorjes1984 · ‎06-02-2009

Hi everyone,

I have some question regarding Multiple WAE deployment in a Data Center:

Case 1

------

We have 1 WAN Router with Multiple WAE Devices, with WCCPv2 configured.

1- How is the load balancing between the WAE devices is applied by WCCPv2.

I did my research and found that it is related to some Hashing algorithm that is based on the source or destination IP of the packet, but i couldn't find any clear explanation about how this algorithm work, can any one advice technically the load balancing is being performed over the multiple WAEs???

2-The DATA cached on 1 WAE, is it replicated on the other WAEs?

Assume that one WAE is disconnected from the group for some reason, does the other WAE's already has the Cached data by that WAE, or they have to learn the traffic again and cached it again.

Case 2

------

We have 2 WAN routers and 1 WAE with the same WCCP group configured on all of them.

1- Does the WAE return the intercepted traffic to the routers according to which originaly redirected the traffic?

All traffic sent from Router#1 to the WAE will be returned to that WAE?? is it possible somehow that the WAE could send traffic back to Router#2 while Router#1 had redirected the traffic to it.

(Assuming both routers are UP and Running)

Regards,

Jorj

jujouber · ‎06-09-2009

Case 1

------

1 - You are correct, by default the load balancing is done using source ip. You can configure the WCCP load balancing based on the hashing algorithm used, either hash assignment or mask assignment. Hash assigns each WAE a specific number of buckets out of a pool of 256. So if 2 WAE's are used each gets assigned 128 bit buckets and the hashing based on src/dst ip is equally shared. Mask assignment allows more granular configuration, the configuration is quite complex.

2 - At this time the WAE's do not share/replicate data between them. If one WAE goes out of service the one that takes over will need to "learn" the cached content again.

Case 2

------

It is possible and this can lead to asymmetric flows which cause performance issues. If multiple routers are used you will need to change the egress method to either wccp negotiated return or generic GRE which forwards the traffic back to the router that initially redirected it over a GRE tunnel.