-
15. Re: cache put stucks forever in a 3 node cluster
Dan:
We use TCP instead of UDP beacuse network environments we run into do not always support UDP (disabled on switchs).
In terms of performance testing, I mentioned in another thread (scalbility consulting). Our performance testing (different tests) are driving loads/request through a load balancer which will spread load (~evenly) among nodes. We definitely see significant performance degration on 3 node cluster comparing 2 node cluster.
I'll try to reproduce it and get the trace.
-
16. Re: cache put stucks forever in a 3 node cluster
Ok, sorry I'm asking you to repeat everything...
Since you have to use TCP, you could also try benchmarking without FRAG2, UFC and MFC. Ignore for now the time it takes to perform the state transfer; I'm pretty sure we can bring state transfer time back to normal by allowing a single sender, we just didn't do it before because we thought UFC/MFC with a small max_credits was a good workaround.
-
17. Re: cache put stucks forever in a 3 node cluster
Dan: I did not forget that.
To my suprise, I was not able to repriduce the issue when I tried it last Friday with the original configuration. Copule weeks ago, I almost hit that issue 3 out 5 runs. I am not sure if some other changes to supress the issue. I'll give another try late.