Correction -- the slow timing this morning was not 623 msec. It was 62.3 msec -- I misread it.
That's still very slow. As I reported in a separate email the delay came entirely from the "DNS+lookup" phase of servicing the request. There was another slow result, at 62.1 msec, all of which came from the "response wait" phase. And in third place, at 24.5 msec, we had "response read". On Wed, May 12, 2021 at 2:20 AM Michael Goulish <mgoul...@redhat.com> wrote: > > I just decided to redo some of my tests with RC1 so I could cast a vote, > and some weird things happened. Possibly good, but weird. > > First, I had to change my PYTHONPATH in one spot to say 2.7 instead of 3.9 > as it was before. Is that correct? > > > > But second, performance on the test I picked -- 'hey' sending HTTP > requests through 1 router to nginx with 10 parallel senders -- has changed > a lot! > > > After seeing the first result I ran it four more times. > > This is exactly the same test I used before -- I have saved the test > scripts and config files, etc, in separate directories so I don't have to > think to re-run the tests. > > > *Executive overview:* > Requests per second have risen by 70%. > The 99% mark is much faster. > The really horrific slowest-measurement did happen once, but it did not > happen on the other four tests. > > > I thought nothing significant should have changed. > What Just Happened? > > > Test: scenario 2 (1 router), 10 senders, TCP adapter > (This is 'hey' sending http requests through one router (TCP) to nginx.) > > RC1 --------------------------------------------------------+--- B*efore: > (0.16.x 0650 DST 04 05 2021)* > | > average : 0.6, 0.6, 0.6, 0.6, 0.6 | *0.6* > | > 99% : 0.5, 0.4, 0.5, 0.4, 0.4 | * 0.7* > | > slowest : 9.1, 623, 3.4, 7.7, 5.3 | *876* > | > requests per second : 30016, 30088, 30124, 30221, 30173 | *17694* > > >
