On Mar 4, 2012, at 18:48 , Jan Lehnardt wrote: > > On Mar 4, 2012, at 18:45 , Bob Dionne wrote: > >> yes, I was surprised by the 30% claim as my numbers showed it only getting >> back to where we were with 1.1.x > > I see ~10% faster than where 1.1.1 was for small docs and ~30% for large docs.
I got curious whether there was a drop-off (or up, I guess) point for the performance improvement between doc size and %-improved. I re-ran the tests for 300, 500 & 700 byte docs respectively and the numbers (updated in the spreadsheet) suggest a ~5% improvement on each iteration. So, looks linear, no sweet spot. Moving on. Cheers Jan -- > > >> I think Bob's suggestion to get to the root code change that caused this >> regression is important as it will help us assess all the other cases this >> testing hasn't even touched yet > > +1 > > Cheers > Jan > -- > > > >> >> On Mar 3, 2012, at 5:25 PM, Bob Dionne wrote: >> >>> I ran some tests, using Bob's latest script. The first versus the second >>> clearly show the regression. The third is the 1.2.x with the patch >>> to couch_os_process reverted and it seems to have no impact. The last has >>> Filipe's latest patch to couch_view_updater discussed in the >>> other thread and it brings the performance back to the 1.1.x level. >>> >>> I'd say that patch is a +1 >>> >>> 1.2.x >>> real 3m3.093s >>> user 0m0.028s >>> sys 0m0.008s >>> ================== >>> 1.1.x >>> real 2m16.609s >>> user 0m0.026s >>> sys 0m0.007s >>> ================= >>> 1.2.x with patch to couch_os_process reverted >>> real 3m7.012s >>> user 0m0.029s >>> sys 0m0.008s >>> ================= >>> 1.2.x with Filipe's katest patch to couch_view_updater >>> real 2m11.038s >>> user 0m0.028s >>> sys 0m0.007s >>> On Feb 28, 2012, at 8:17 AM, Jason Smith wrote: >>> >>>> Forgive the clean new thread. Hopefully it will not remain so. >>>> >>>> If you can, would you please clone https://github.com/jhs/slow_couchdb >>>> >>>> And build whatever Erlangs and CouchDB checkouts you see fit, and run >>>> the test. For example: >>>> >>>> docs=500000 ./bench.sh small_doc.tpl >>>> >>>> That should run the test and, God willing, upload the results to a >>>> couch in the cloud. We should be able to use that information to >>>> identify who you are, whether you are on SSD, what Erlang and Couch >>>> build, and how fast it ran. Modulo bugs. >>> >> >
