Sounds good, Dmitri. I'll work up something on a fork in the next few days and ping you. We can go from there.
Jim On Mon, Feb 29, 2016 at 2:33 PM, Dmitri <dmitri.sotni...@gmail.com> wrote: > Hi Jim, > > I'm the author of Luminus, and I'd love tow work with you to tune up the > performance. Feel free to ping me via email or on GitHub. > > > On Sunday, February 28, 2016 at 8:17:13 PM UTC-5, Jim Crossley wrote: >> >> I just tried a few experiments and realized the :dispatch? option is >> broken in the latest Immutant release. :( >> >> This is a result of some changes we made to better support WebSockets. >> >> We were already hoping to get a release out this week, so we'll add that >> to the list of fixes. >> >> Sorry about that, >> Jim >> >> On Sunday, February 28, 2016 at 1:32:09 PM UTC-5, Jim Crossley wrote: >>> >>> Hi, >>> >>> Luminus uses Immutant, which uses Undertow, so it should be possible >>> to tune the Luminus app to approach the performance of the TechEmpower >>> Undertow app. The relevant options to immutant.web/run [1] are >>> :dispatch?, :io-threads, and :worker-threads. >>> >>> The Undertow app sets IO threads here [2] and worker threads here [3], >>> so you can easily set those same values in the Luminus app. >>> >>> But the real performance bump will come from the :dispatch? option. By >>> default, :dispatch? is true, i.e. requests handled by threads in the >>> IO pool are dispatched to a thread in the worker pool. For >>> compute-bound tasks like the JSON serialization benchmark, that >>> context switch is far more expensive than just having the IO thread >>> return the response. So you'd want to set :dispatch? to false in that >>> case. >>> >>> The tricky bit here is that the Luminus app defines all its routes in >>> a single handler [4]. But for the more io-bound tasks, e.g. >>> DbSqlHandler [5], you'll want the default true value of :dispatch?. So >>> in order to get the best results for all the benchmarks, you'll need >>> to re-organize that app to run two handlers: one that dispatches and >>> one that doesn't. In Immutant, the simplest way to do this is to >>> distinguish each handler with a unique :path option. >>> >>> I'm happy to assist whoever is maintaining that app with the tuning. >>> >>> Thanks, >>> Jim >>> >>> [1] >>> http://immutant.org/documentation/current/apidoc/immutant.web.html#var-run >>> [2] >>> https://github.com/TechEmpower/FrameworkBenchmarks/blob/master/frameworks/Java/undertow/src/main/java/hello/HelloWebServer.java#L122 >>> [3] >>> https://github.com/TechEmpower/FrameworkBenchmarks/blob/master/frameworks/Java/undertow/src/main/java/hello/HelloWebServer.java#L160 >>> [4] >>> https://github.com/TechEmpower/FrameworkBenchmarks/blob/master/frameworks/Clojure/luminus/hello/src/hello/routes/home.clj#L44 >>> [5] >>> https://github.com/TechEmpower/FrameworkBenchmarks/blob/master/frameworks/Java/undertow/src/main/java/hello/DbSqlHandler.java#L38 >>> >>> On Sat, Feb 27, 2016 at 9:40 PM, gvim <gvi...@gmail.com> wrote: >>> > In the latest round of Techempower benchmarks: >>> > >>> > >>> > https://www.techempower.com/benchmarks/#section=data-r12&hw=peak&test=json >>> > >>> > ... I was surprised to find Luminus performing no better than Hapi >>> > (Node) >>> > and significantly worse than Java frameworks. Figures are >>> > requests/second: >>> > >>> > FORTUNES >>> > - Hapi: 1.9 >>> > - Luminus: 0.9 >>> > - Gemini: 55.5 >>> > >>> > JSON SERIALISATION >>> > - Hapi: 0.3 (Raw db) >>> > - Luminus: 0.8 >>> > - Rapidoid: 78.4 >>> > >>> > >>> > SINGLE QUERY >>> > - Hapi: 2.9 >>> > - Luminus: 8.7 >>> > - Gemini: 75.8 >>> > >>> > MULTI-QUERY >>> > - Hapi: 33.0 >>> > - Luminus: 20.4 >>> > - Dropwizard: 65.8 >>> > >>> > DATA UPDATES >>> > - Hapi: 20.9 >>> > - Luminus: 20.0 >>> > - Ninja: 54.7 >>> > >>> > PLAINTEXT >>> > - Hapi: 0.7 >>> > - Luminus: 0.0 >>> > - Rapidoid: 100.0 >>> > >>> > >>> > >>> > Any ideas? >>> > >>> > gvim >>> > >>> > >>> > >>> > >>> > >>> > >>> > >>> > >>> > -- >>> > You received this message because you are subscribed to the Google >>> > Groups "Clojure" group. >>> > To post to this group, send email to clo...@googlegroups.com >>> > Note that posts from new members are moderated - please be patient with >>> > your >>> > first post. >>> > To unsubscribe from this group, send email to >>> > clojure+u...@googlegroups.com >>> > For more options, visit this group at >>> > http://groups.google.com/group/clojure?hl=en >>> > --- You received this message because you are subscribed to the Google >>> > Groups "Clojure" group. >>> > To unsubscribe from this group and stop receiving emails from it, send >>> > an >>> > email to clojure+u...@googlegroups.com. >>> > For more options, visit https://groups.google.com/d/optout. > > -- > You received this message because you are subscribed to the Google > Groups "Clojure" group. > To post to this group, send email to clojure@googlegroups.com > Note that posts from new members are moderated - please be patient with your > first post. > To unsubscribe from this group, send email to > clojure+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/clojure?hl=en > --- > You received this message because you are subscribed to the Google Groups > "Clojure" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to clojure+unsubscr...@googlegroups.com. > > For more options, visit https://groups.google.com/d/optout. -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
