04-04-2003 05:26 AM - edited 03-02-2019 06:24 AM
Anyone with any LD experience, assistance is greatly appreciated!
Heres the scenario. We have a local director load balancing between two applications servers. The sessions to these servers, once established, are very long and can last most of the day. If one of the two servers goes down, the sessions are directed to the second server. Because of the length of the sessions, once the first server comes back online, all of the established sessions are still with the second server and will not move back until the sessions end and have to reconnect. This seems to have overloaded the second server's CPU on many occasions. This local director was basically an "out of the box" setup and I'm not sure if it was configured to have this function. Is there anyway through configuration that we can have the LD start gracefully switching sessions back to the first server once it is back online to offload the second one?
04-10-2003 06:27 AM
Sorry but this is not possible (I don't even think with CSS or
any thing I have ever heard of) Given TCP and everything I
know about it, I don't think could ever be possible, unless
the load balancer was handling layer 4 and down and doing
all negotiation layers 5 to 7.
Tweaking the predicator (method of determining who gets
NEW connections) will allow you to make sure any NEW
sessions made don't go to the second server (because it
will already have some established connections), but
any EXISTING connections will remain on the server which
has been up for the entire time of the connection.
The Predicator I use is "Least Conns" so the server with the
least ACTIVE connections always gets the new requests for
service off the VIP. If you run Least conns as well, you can be
pretty sure to evenly distribute your connections amoungst the
number of "real" servers in your pool. You should convince your
management that more "real" servers are needed to protect
against this crash that the Local Director (LB) was never invented
for. Say if you had 8 servers instead of 2, then you would not crash
the remaining servers if one failed.
04-10-2003 07:07 AM
Thanks for the help!
Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: