2020-09-29 00:25:56 UTC - Dominic Kim: What is the value configured for your `userMemory`? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339156004900?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:26:21 UTC - Dominic Kim: Generally, memory is the dominant factor in invokers. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339181005100?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:33:38 UTC - Kostis Kaffes: It is large, 104GB. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339618005300?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:34:34 UTC - Dominic Kim: ok then memory would not be the one. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339674005500?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:34:39 UTC - Dominic Kim: What is your log level? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339679005700?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:34:55 UTC - Kostis Kaffes: info https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339695005900?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:35:09 UTC - Dominic Kim: I observed the `WARN` level outperform the `INFO` level setup. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339709006100?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:35:15 UTC - Kostis Kaffes: I see. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339715006300?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:35:35 UTC - Dominic Kim: Let me recall my memory. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339735006500?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:36:00 UTC - Kostis Kaffes: What is the highest invocation rate you have achieved with a single Controller assuming unlimited Invokers? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339760006700?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:39:42 UTC - Dominic Kim: There are some differences in the code base as I performed this long time back, I found this result. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339982006900?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:39:46 UTC - Dominic Kim: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601339986007100?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:40:10 UTC - Kostis Kaffes: Thanks! I am not using a very recent version actually as well. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340010007500?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:40:10 UTC - Dominic Kim: So when there are 1 contoller with 4 invokers, I got around 4,000 RPS. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340010007700?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:40:21 UTC - Kostis Kaffes: This is reasonable. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340021007900?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:40:25 UTC - Kostis Kaffes: Thanks for all the help! +1 : Dominic Kim https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340025008100?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:41:01 UTC - Dominic Kim: How many virtual clients did you use? I think that would be also one of the dominant factors. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340061008400?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:41:43 UTC - Kostis Kaffes: 12 multi-threaded clients, I don’t think that is the bottleneck. white_check_mark : Dominic Kim https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340103008600?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:41:57 UTC - Dominic Kim: And JFYI, I used virtual machines with 8 cores and 16GB memory each. +1 : Kostis Kaffes https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340117008800?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:43:58 UTC - Dominic Kim: It seems I utilized 990 virtual clients (threads in your case I suppose). https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340238009200?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 00:44:39 UTC - Kostis Kaffes: I will re-run with log level WARN. +1 : Dominic Kim https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601340279009400?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 01:04:46 UTC - Kostis Kaffes: It varies between 3500-3800 RPS, quite close to the 4000 you achieved. white_check_mark : Dominic Kim https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601341486009700?thread_ts=1601186283.000800&cid=C3TPCAQG1 ---- 2020-09-29 22:51:00 UTC - Brendan Doyle: Is there any documentation on the performance improvements of using docker-runc on invokers over docker? How much of an improvement can be expected? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601419860011200?thread_ts=1601419860.011200&cid=C3TPCAQG1 ---- 2020-09-29 23:58:57 UTC - Brendan Doyle: also does it still hold true that the docker server version on the host must strictly match the docker client version in the container to use runc? We're attempting to upgrade from 1.13 to 18.06 which we see was done over a year ago in the main repo https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1601423937011300?thread_ts=1601419860.011200&cid=C3TPCAQG1 ----
