I wouldn't worry too much about the benchmarks. Micro benchmarks can sometimes be tricky to design and interpret. It appears that Throughput is calculated over all threads (Fork=3) whereas AverageTime is per op, single thread so it's not the actual inverse of each other. Also for throughput, the actual boost here is (598/580) 3% not 10%. The variance wasn't published but if it's anything like the 3% for Average Time, then the boost is marginal. I don't think it will be perceptible unless it's over a huge dataset but even then there are so many other things to contend with. The throughput benchmark points to how quickly that performance boost dissipates.
In any case, I think we can all agree that in general not having a method call overhead is going to be slightly faster. As long as the code changes are sound, let's just take what we can get. [ Full content available at: https://github.com/apache/tinkerpop/pull/1001 ] This message was relayed via gitbox.apache.org for [email protected]
