topic Re: Source IP stickyness question in Application Networking

Source IP stickyness question

quesne02 — Tue, 28 Apr 2009 11:30:35 GMT

In the parameters to be set for stickyness based on IP, there is a mask to be configured.

Can someone explain the behaviour of changing the length of this mask, say 255.255.255.0 versus 255.255.255.128 or any other permutation.

Also, I would like to understand the link with the predictor. How does the 2 really react and is there some permutation that work best with source IP stickyness.

Thanks

Re: Source IP stickyness question

dario.didio — Tue, 28 Apr 2009 11:58:52 GMT

Hello,

I think this can be best explained using an example.

3 hosts connect to your VIP address:

host1 - 192.168.0.5

host2 - 192.168.0.200

host3 - 172.16.0.20

imagine you use source sticky with a netmask of 255.255.255.0

host1 connects to the VIP and a check for is done if he already exists in the sticky table. this is not the case so the predictor is used to balance. he goes to server 1, and a sticky entry is added to the sticky database (I don't know the way this is done internally, but the result will be that any host with source address 192.168.0.0/24 (<-- netmask) is sticked to server1.

Host 2 arrives and the same process is done; a lookup in the sticky database results in a match, because host1 its source matches 192.168.0.0/24, and he is also send to server1.

Host3 does the same thing, again a lookup in the sticky database, no entry found, balance using predictor, send to server2 and added to sticky database.

Imagine if we used a netmask of 255.255.255.128, in this case host1 and host2 would not match each other and would be balanced individually.

Concerning the predictor, this is only used when no sticky entry exists for a certain connection. I think it is the same question you can pose yourself when you need to choose a predictor with or without stickyness.

HTH,

Dario

Re: Source IP stickyness question

quesne02 — Tue, 28 Apr 2009 12:05:15 GMT

So the mask really applies close to a subnet mask to the incoming client coming in.

All clients within this mask will go to the same server based on internal calculation.

I now also understand the relationship with the predictor.

Thanks