http://albany.eng.hst.ams2.redhat.com/job/narayana-benchmarks/52/

The rollbacks are most likely due to transaction timeouts (even though I set the threshold for failure to not more than 10 rollbacks).

It is possible that JOTM just cannot cope with the number of threads we using to produce the benchmark. I have tried to reproduce the failures with a standalone junit test and using JMH on my laptop without success.

I am going to remove JOTM from the comparison matrix (see job narayana-benchmarks) until this JIRA is resolved.