> On April 30, 2015, 5:29 p.m., Bill Farner wrote: > > src/jmh/java/org/apache/aurora/benchmark/StatusUpdateBenchmark.java, line > > 103 > > <https://reviews.apache.org/r/33608/diff/4/?file=945995#file945995line103> > > > > I should have asked the first time around - what's the thought process > > behind including this? Given that this is a benchmark, it seems only to > > place a ceiling on throughput.
My goal here was not to micro-benchmark the status update processing path, but to show how to achieve high throughput against an expensive resource. So, although it's obviously the case today that the latency bounds throughput, this benchmark serves to demonstrate that _even in the face of degraded storage latency_ we can have high throughput. With my upcoming changes, that is. :) - Ben ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/33608/#review82163 ----------------------------------------------------------- On April 30, 2015, 12:35 a.m., Ben Mahler wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/33608/ > ----------------------------------------------------------- > > (Updated April 30, 2015, 12:35 a.m.) > > > Review request for Aurora and Maxim Khutornenko. > > > Bugs: AURORA-1283 > https://issues.apache.org/jira/browse/AURORA-1283 > > > Repository: aurora > > > Description > ------- > > In order to justify doing asynchronous batch acknowledgements and to better > understand status update throughput, this introduces a benchmark. > > Note that this assumes that status update processing is not synchronous, so > that the benchmark doesn't need to be updated for AURORA-1228. > > > Diffs > ----- > > src/jmh/java/org/apache/aurora/benchmark/StatusUpdateBenchmark.java > PRE-CREATION > src/jmh/java/org/apache/aurora/benchmark/fakes/FakeOfferManager.java > PRE-CREATION > src/jmh/java/org/apache/aurora/benchmark/fakes/FakeSchedulerDriver.java > PRE-CREATION > src/main/java/org/apache/aurora/scheduler/UserTaskLauncher.java > c54619f7cd617b48069363173dcf63b6254b4095 > src/main/java/org/apache/aurora/scheduler/mesos/SchedulerDriverModule.java > d7d659bb13f085ff06291ef0033572f8bdf29874 > > Diff: https://reviews.apache.org/r/33608/diff/ > > > Testing > ------- > > Ran the benchmarks against the existing code and some pending code I have for > AURORA-1228 to demonstrate the improvement. > > The end to end tests are broken, appears to be unrelated to my change from > what I can tell. > > > Thanks, > > Ben Mahler > >
