-
1. Re: Improving request distribution for stateless applications.
jfclere Feb 28, 2014 11:04 AM (in response to edwinb1)So you cluster nodes are grouped in domains?
-
2. Re: Improving request distribution for stateless applications.
edwinb1 Feb 28, 2014 2:22 PM (in response to jfclere)Yes, my environment hosts various applications, with each under a distinct LBGroup and several available nodes. Some applications/LBGroups are stateful and are configured with session stickiness and a variable load-factor setting, but the few that are stateless can really benefit from a more random request distribution across available nodes.
My initial testing against a simple test application after 10K requests seems to demonstrate a fairly well balanced outcome, but I wanted to confirm that LBGroup boundaries are preserved in light of MODCLUSTER-178. I've not yet seen any problematic requests but I've only just begun testing this.
Thank you.
-
3. Re: Improving request distribution for stateless applications.
jfclere Mar 3, 2014 5:21 AM (in response to edwinb1)1 of 1 people found this helpfulThe stickiness of sessions is defined in the balancer so it can't be changed at the webapp level.
The rule for fail-over is to fail-over to a node of the same domain and then to another domain.
MODCLUSTER-178 has been implemented for the moment.
-
4. Re: Improving request distribution for stateless applications.
edwinb1 Mar 3, 2014 7:43 AM (in response to jfclere)If I set the Balancer property for each of my webapps to a distinct value, in addition to LBGroup and JVMRoute, can that work or would I have to setup another mod_cluster instance to host stateless webapps?
If I can avoid changing session stickiness and simply install a static load-factor provider for stateless webapps, would that yield a random reqeust distribution?
-
5. Re: Improving request distribution for stateless applications.
jfclere Mar 7, 2014 10:07 AM (in response to edwinb1)Changing the balancer property either creates a new balancer or update an existing one so that won't help, if you group the stateless webapps on some nodes having another mod_cluster instance for those nodes could make sense (another balancer too also in that case). The stickiness is on node not on applications so you have to separe sticky webapp and stateless ones.