[infinispan-issues] [JBoss JIRA] (ISPN-2625) Verify Cross-site transfer performace

Verify Cross-site transfer performace ------------------------------------- Key: ISPN-2625 URL: https://issues.jboss.org/browse/ISPN-2625 Project: Infinispan Issue Type: Task Components: Cross-Site Replication Affects Versions: 5.2.0.Beta5 Reporter: Radim Vansa Assignee: Mircea Markus Attachments: jgroups-lon-nyc-sfo-belaban.xml, jgroups-site.xml, lon-tx-optimistic-xs-to-nyc-async.xml, nyc-tx-optimistic-xs-from-lon.xml The performance of asynchronous cross-site transfer is questionable. In QA lab we got 2x-4x slower reads (GET) than in no-XS scenario and write (PUT) performance equivalent to those of synchronous XS. While the code cannot be reliably profiled with instrumentation, trace logs also radically slow down the test we have not identified any particular bug that should cause this issue. Therefore, I'd like to ask someone from the core developers to try to reproduce such test results and provide his test description together with achieved results. We have used three-node cluster in distributed mode with optimistic transactions, replicating to another three-node cluster with the same configuration. Physically the nodes were in the same cluster without any delay element. The logical clusters used different multicast addresses in their JGroups configuration. Each node started 10 threads which issued PUT/GET requests in a loop with 1:4 ratio. Each request was wrapped in single transaction. The test was benchmarking the average duration of each operation during 10 minute period preceded by 3 minute warmup with identical configuration but not accounted into performance results. JGroups configurations as well as Infinispan configuration will be enclosed.