Some performance regression tests have very variable execution time.
An example is 04dd_join4kl.ecl.
On a 1 slave (slavesPerNode=1, channelsPerSlave=1) vanilla (6.4) setup, the times vary from a low ~2.5seconds to over 60seconds on my machine.
Turning off RHFnofragment [ via #option ('heapflags', 0) ] - produces consistent timings of ~2.5seconds
|Avoid allocating rows from multiple blocks||Resolved||Unassigned|