Create an ultra high performance distributed transaction coordinator to handle infinispan, JMS, websevice(with compensation logic), file system resources and database transactions. Current XA implementations cause high performace degradation.
Transaction logs should be stored in massively parallelized sql/file backing stores(Use an infinispan write through cache region to store tx logs?) on the nodes.
Transaction manager should be able to work only with a webcontainer lie Tomcat (without a real application server).
Transaction coordinator should use blocking free(As much as possible. Use profiler like JProfiler or others heavily to detect locking/blocking issues) architecture in the source code to allow ultra high transaction counts (thousands/sec per server). It is necessary for new distributed cache architectures.