I have an ACE deployment with 4 contexts. In all the 4 contexts I have configured for Roundrobin load balance with 2 servers and sticky source IP. This works well.
But sometimes in one of the context, the Round robin LB not happens to appear. All the client requests from different sources gets stick to only one server. When I manually clear the sticky database entries for this particular group, the Round robin LB starts working and both servers gets hits.
Is this relates to some bug or configuration issues?
Are you not seeing any connections to other Rserver or the connections are not evenly distributed?
With Sticky+roundrobin It not necessary to see even number of connections distributed between rservers.
If users stuck to a rserver are more frequent (they connect again before sticky timeout expires) then the other users, then you will see more connections on that rserver.
Thanks for the reply.
the users are stuck to only one server untill I manually clear the sticky entries and Later it is doing Round Robin.
Thanks for the replies.
So you mean to say that eventhough the Round robin predictor is configured with source IP based sticky, when different clients try to hit the VIP, they wil be not be load balanced by round robin, but will stick to only one server.
If so, I need a round robin with sticky, what type of stickiness should I configure?
Can you please send me a link of the behaviour of Round robin predictor with sticky, so that it will be helpful for me to confirm my customer?
What We want to say is
If a client is stuck to a server then he will remain stuck there if he keeps coming back within the sticky time out.
Lets assume you are using /32 mask sticky group and one hr sticky timeout.
If client from 220.127.116.11 hits the vip and he gets stuck to Server 1 then he will remain there unless he seize to hit VIP for an hour.
If another client from 18.104.22.168 hits the vip he will be given Server 2 by ACE (due to round robin).
ALL new clients (IPs not in sticky database) will get real server on roundrobin basis.
If a client hits a VIP and his IP is already in the Sticky database then loadbalancing logic (roundrobin in your case) will not come into play. These clients will simply be sent to the servers selected previously by the ACE.
Due to this behavior its very normal to see uneven connections between reals in a serverfarm.
Sorry, I think I have not explained my problem correctly.
If we have two clients one at 22.214.171.124 and other at 126.96.36.199, both get stuck to only one rserver even when the other is active and free. This happens untill I clear the sticky database.
when I clear the sticky database, 188.8.131.52 goes to rserver1 and 184.108.40.206 goes to rserver2 and 220.127.116.11 goes to rserver1 and 18.104.22.168 goes to rserver2.
Why stucking to only one rserver is happening?
Are you looking at the "current conns"/"Total Conns" in "sh serverfarm detail"?
I would recommend clearing the serverfarm statistics (clear serverfarm
connections" column and not the "Current conns" column.
Round robin algo doesn't care about the current number of connections, it only takes into account the total number of connections sent. Current connections can be more on 1 server vs other just because of HTTP1.1 persistence (where multiple GETS are sent on a single TCP connection).
If you want to loadbalance on "current connection" basis then change the predictor to
Syed Iftekhar Ahmed
We observed in two ways, one from the real servers standpoint and the other from the show server farm detail. I tried even clearing the show server farm detail statistics.
I will try with the Least connections predictor. How much it will benefit me when I disable the HTTP persistence from the ACE configuration for using the round robin.
Even if you disable HTTP persistence then also the clients that are stuck to a real will send conns to the same real.
In order to see most even current conns you will have to disable Persistence & Stickiness :)
Could you post the sh serverefarm detail output.
There are lots of Connections going to both Rservers. Its not the case where your other servers are not servicing at all.
Its important to understand that when using stickiness, no real even load
balancing is going to happen since we are sticking requests.
Everything was going normal all these days, since the real servers are into production more than 10 months. Now suddenly , we started getting hits to only one server. then when we cleared the sticky database both the servers started to receive hits.
That was my concern.