CSS Balance srcip = sticky source ip?

[cypher.jay]

Hi

Is the balance method "srcip" the same as the general method known as "sticky source ip"? Whereby, the same source ip will be directed to the same destination server by the CSS?

Thanks in advance.

[Gilles Dufour]

it's not exactly the same.

With balance srcip, the CSS uses an algorithm that guarantees that the same source is sent to the same server.

No sticky table is being used.

Therefore, if server A is down, and source S comes in, it is sent to server B.

However, when server A comes back online, it is possible for the CSS to select source A instead of B.

With all servers up, it is however guaranteed that the CSS will select the same server for the same source.

'advanced-balance sticky-srcip' is using a sticky table and it will store the server being used for each source. Even if the servers are flapping, the same source is guaranteed to go to the same server.

The only disadvantage is that the CSS needs to maintain a sticky-table.

Regards,

Gilles.

[cypher.jay]

Hi Gilles,

Thanks for getting back to me on this. So it seems to me that the preferred balance would be srcip instead of advanced-balance sticky-srcip.

Since it seems that with srcip method, the CSS is smart enough to not just blindly directs to the tied-in server (if the server happens to flap) but will redirect the source to another available server.

If so, why would anyone prefer using sticky-srcip then? A table is maintained, and the same source will always be directed to the same server, eventhough it's flapping up/down?

Thanks.

[Gilles Dufour]

sorry, but it looks like my explanation was not clear enough.

Let's take the following scenario.

We have 2 servers - A & B.

Initially A is down.

A client C connects to the vip and is therefore loadbalanced to B based on the srcip algorithm.

After a while, A comes alive.

C open a new connections with the VIP and this time the algorithm selects A.

So stickyness did not work.

With 'advanced-balance srcip' the first connection will go to B and a sticky entry will be created.

When the 2nd connection comes in with both A & B alive, the CSS will forward the connection to B based on the sticky table.

If B was down, the CSS would forward the connection to A and update the sticky table.

So, the advanced method gives you a better guarantee of stickyness.

As I said, the only problem is the sticky table.

Actually this is most of the time not a problem, but sometimes you could end up with the table being full and stickyness start failing. There is solutions to this problem so.

I hope this makes more sense like this.

Gilles.

With balance srcip