We busy with a project to upgrade from Infinispan 14.0.1 to 14.0.20 but have found that our distributed (ASYNC) caches are signaficantly slower in the new version.
In 14.0.1 we got an average write time of 0.5ms and a read time of 0.4ms while on 14.0.20 write times were down to 0.2ms but read times jumped to 1ms. We have tried using all the version between 14.0.1 and 14.0.20 and found that the slow response problem started with version 14.0.16.
We have a three node cluster configured with a MPING jgroup setup. We have further opserved that if we only run with one node the response issue disappears and both read and write completes in 0.2ms.
Any suggestions on how to solve this issue?
After liaising with the Infinispan team, the issue persisted on the hotrod client and not the Infinispan Cluster.
We were incorrectly specifying the the client intelligence to BASIC.
quarkus.infinispan-client.client-intelligence=BASICThe basic setting should only be used for standalone machines and/or local testing.
Setting the intelligence level to 3 resolved the issue.