Channels
#server
- o
orangy
02/10/2017, 8:07 PMdid you check #ktor or #wasabi? I wonder why make a new one. - d
danneu
02/10/2017, 8:10 PMyeah, i was familiar with the existing kotlin web servers. more of a curiosity than anything. kotlin is actually the third language i've implemented this little stack in. it's a hobby of mine. but the kotlin iteration was the most promising, imo - d
danneu
02/10/2017, 8:12 PMi could've built a set of smaller projects to learn the language, but that's what a web framework kinda is - a bunch of little minigames - o
orangy
02/10/2017, 8:12 PMthat I totally understand! so much fun digging into web fundamentals 🙂 - o
orangy
02/10/2017, 8:13 PMthe only thing that saddens me a bit, that there a 2 people working on ktor, ~2 on wasabi, one on kog, etc, etc. If we only could join our forces… - d
danneu
02/10/2017, 8:16 PMyeah. when i started kog i had no idea i'd be playing with it beyond that weekend - o
orangy
02/25/2017, 10:06 PMI’ve set up JMH benchmarks for various servers https://github.com/orangy/http-benchmarks Based on prior work at https://github.com/rocketraman/kotlin-web-hello-world, but with automated measurement of all servers at once.👍 5 - r
rocketraman
02/25/2017, 11:33 PMNice! - r
rocketraman
02/26/2017, 12:03 AM@orangy Looks like its missing a repo definition for ktor-netty and ktor-jetty snapshot - r
rocketraman
02/26/2017, 12:03 AM[ERROR] Failed to execute goal on project http-benchmarks: Could not resolve dependencies for project org.kotlin.community:http-benchmarks:jar:0.1-SNAPSHOT: The following artifacts could not be resolved: org.jetbrains.ktor:ktor-netty:jar:0.3.0-SNAPSHOT, org.jetbrains.ktor:ktor-jetty:jar:0.3.0-SNAPSHOT: Failure to find org.jetbrains.ktor:ktor-netty:jar:0.3.0-SNAPSHOT in <https://jitpack.io> was cached in the local repository, resolution will not be reattempted until the update interval of jitpack has elapsed or updates are forced -> [Help 1] - o
orangy
02/26/2017, 6:34 AMOuch, right, it’s from
branch I was working on with performance improvements.coroutines - k
kenkyee
02/26/2017, 12:20 PMWhat would also be interesting is memory usage. E.g. if it uses coroutines or reactive, I'd expect it to use less memory than one that uses old style threads... - o
orangy
02/26/2017, 1:07 PMDefine “memory usage”? - o
orangy
02/26/2017, 1:08 PMIs it allocation rate, retained average memory, GC time, etc? - o
orangy
02/26/2017, 1:31 PMFound http://there4.co/hexagon/, but it won’t run with newer Jetty… - o
orangy
02/26/2017, 1:32 PM@jaguililla ^^ - o
orangy
02/26/2017, 9:17 PMAdded a number of frameworks, like Grizzly, Colossus, NinjaFramework - o
orangy
02/26/2017, 9:18 PMInteracting with Scala framework is a pain… (see github.com/orangy/http-benchmarks/blob/master/src/org/kotlin/community/http/benchmarks/colossus/ColossusBenchmark.kt#L24-L40)😱 1 - k
kenkyee
02/27/2017, 1:58 PM"memory usage" would be average retained memory...I.e., what would cause a lot of swapping if the underlying core is not efficient - k
kenkyee
02/27/2017, 1:59 PMGC time probably directly correlates with allocation rate - o
orangy
02/27/2017, 2:16 PMI’m afraid I don’t have enough facility in JMH to measure that. Should probably ask Shipilev 🙂 - o
orangy
02/27/2017, 2:19 PMAh, may be GC profiler would be useful, let me try… - o
orangy
02/27/2017, 2:25 PMokay, it works, but obviously skews performance benchmarks. So it needs to be run twice. - o
orangy
02/27/2017, 2:43 PMPreliminary resultsCopy codeo.k.c.h.b.undertow.UndertowBenchmark.hello:·gc.alloc.rate.norm thrpt 5 17915.800 ± 2473.661 B/op o.k.c.h.b.jetty.JettyBenchmark.hello:·gc.alloc.rate.norm thrpt 5 18272.728 ± 1202.523 B/op o.k.c.h.b.vertx.VertxBenchmark.hello:·gc.alloc.rate.norm thrpt 5 18375.959 ± 5136.971 B/op o.k.c.h.b.netty.NettyBenchmark.hello:·gc.alloc.rate.norm thrpt 5 19039.439 ± 324.563 B/op o.k.c.h.b.colossus.ColossusBenchmark.hello:·gc.alloc.rate.norm thrpt 5 20534.041 ± 630.114 B/op o.k.c.h.b.spark.SparkBenchmark.hello:·gc.alloc.rate.norm thrpt 5 20725.427 ± 483.491 B/op o.k.c.h.b.ktor.KtorJettyBenchmark.hello:·gc.alloc.rate.norm thrpt 5 24808.862 ± 16120.989 B/op o.k.c.h.b.ktor.KtorNettyBenchmark.hello:·gc.alloc.rate.norm thrpt 5 26347.204 ± 19105.626 B/op o.k.c.h.b.nanohttpd.NanoHttpBenchmark.hello:·gc.alloc.rate.norm thrpt 5 31304.811 ± 10778.326 B/op o.k.c.h.b.akka.AkkaHttpBenchmark.hello:·gc.alloc.rate.norm thrpt 5 31434.074 ± 984.795 B/op o.k.c.h.b.grizzly.GrizzlyBenchmark.hello:·gc.alloc.rate.norm thrpt 5 43390.309 ± 55646.962 B/op o.k.c.h.b.ninjaframework.NinjaBenchmark.hello:·gc.alloc.rate.norm thrpt 5 235810.785 ± 466148.074 B/op o.k.c.h.b.fluenthttp.FluentHttpBenchmark.hello:·gc.alloc.rate.norm thrpt 5 2798131.227 ± 5984986.241 B/op👍 3 - m
miha-x64
02/27/2017, 2:47 PMWhat about Spring? 🧌 - o
orangy
02/27/2017, 2:47 PMPlease contribute!😞 3 - o
orangy
02/27/2017, 2:48 PMOr may be ask @sdeleuze to do it 🙂 - s
sdeleuze
02/27/2017, 4:23 PMI will be happy to contribute Reactor and Spring benchmarks, but the from a quick look, I don't think the current benchmark give a useful indication of the performance you will have in a real web application, ie. just benchmarking returning a String from GET endpoint is maybe useful to check you have no a basic performance issue, but I would not use that to compare Web Frameworks. Something implying request path parsing, content negotiation and JSON serialization of a data coming with some latency and more that a single route defined would give a much more interesting overview of web framework performances.➕ 4 - o
orangy
02/27/2017, 4:33 PMOf course, this is something I understand very well, but it’s just a start. I want to make sure I have basic benchmarks setup right, before advancing to downloading files, and then to adding sync and async load. - o
orangy
02/27/2017, 4:33 PMRight now performance results agree very well with TechEmpower benchmarks (relatively to each other), so I assume it’s basically ok.