Come on, I know that my two new more complex topics usually get few answers.
However, we have just at this a little stress and any support is very welcome. Don't any of you know of a really good LB expert who can assess the advantages of the individual approaches? Or briefly describe his own approach to me.
2 of 2 people found this helpful
LB serves two functions in my mind, obviously, balancing the load, the second is HA Redundancy. With your current configuration, regardless of which you use, you are only handling the first of the two advantages of an LB.....but, if the node you are on goes down, then you are balanced to the other node and you loose your session and are forced to log back on....so, let me propose an alternate configuration you may not have thought of.
Cluster your Tomcats so that sessions are replicated between the two of them. Then turn off session affinity at the LB side (because it's not needed at that point) and let it use least connections to ballance you. In this scenario you have proper ballancing, no stickyness, AND HA redundancy.
The only con to this configuration is CPU/Memory overhead on keeping the sessions sync'd, which only becomes problematic in larger configurations with > 4 MT's....but if you truly only have 2 as you indicate, it shouldn't be problematic, and it solves your situation entirely.
1 of 1 people found this helpful
interesting approach, I know only from stateless applications. But the midtier can't do tha, at least that's what BMC writes everywhere.
By the way, we have more than 2 MT, I just wanted to keep the example as simple as possible. What does your approach to 4+ MT look like if your alternative idea doesn't work anymore?
3 of 4 people found this helpful
FYI, documentation for the clustering is here
in case you cared
So...if you don't want to/can't go with a Tomcat cluster, thus enabling seamless fail over and no need for sticky bits, I would personally go with option 1, not for any particular reason, but simply because I like the simplicity of it and don't care if the same user is on multiple mid-tier machines....truly that doesn't matter to me. Docs I found while doing some research for you, in case you cared to look, but you likely don't need to
3 of 3 people found this helpful
a Tomcat cluster is no problem for us, we had to run it for many months in parallel. So that in the first step always the LB sticky grabs, should there be problems the Tomcat cluster grabs. It gets complicated with RSSO, something doesn't seem to work with the Tomcat cluster and RSSO integration. This is my other thread, unfortunately without much interest.
Here I want to know how others have configured their LB with sticky.
Your tip without sticky, contrary to the BMC docu, but worth considering
Thanks in advance