2014/1/6 Yo-An Lin <cornelius.howl@xxxxxxxxx> > Here is the benchmark details, you can find the benchmark code below: > > https://github.com/c9s/router-benchmark > Hi, Well, you still use a full LAMP-stack and HTTP-benchmark-tool for benchmarking. That was actually _the_ point, which makes this benchmark useless for me... Beside this: You should increase the routes you benchmark and introduce more variation (pattern-matching, matching first/last route and so on), because different implementations often differ in different situations. > > on intel xeon 16 cores machine: > > symfony/routing only got 4873.17 #/sec while phux got 16777.12 #/sec > > on intel core i5 machine: > > symfony/routing only got around 200-300 #/sec while phux got around > 1600-1800 #/sec > > > > Also the pure php dispatch benchmark: > > cat dispatch.txt > n=10000 > Runing symfony/routing - . 6664.4093423989/s > Runing phux extension - . 16501.898328291/s > > Rate Mem phux extension symfony/routing > phux extension 16K/s 5M -- -40% > symfony/routing 6K/s 786K 247% -- > > > ================================== Bar Chart > ================================== > > phux extension 16K/s | > ████████████████████████████████████████████████████████████ | > symfony/routing 6K/s | ████████████████████████ > | > > > ============================== System Information > ============================== > > PHP Version: 5.5.6 > CPU Brand String: Intel(R) Core(TM) i5-2400S CPU @ 2.50GHz > > > > > > > On 2014/1/6, at 下午12:36, Yo-An lin <cornelius.howl@xxxxxxxxx> wrote: > > Yeah I know, the benchmark is pretty rough. Because its a two day work. :p > > The details will come up later. > > [Sent from iPhone] > > Sebastian Krebs <krebs.seb@xxxxxxxxx> 於 2014/1/6 7:54 寫道： > > > > > 2014/1/5 Lin Yo-An <cornelius.howl@xxxxxxxxx> > >> Hi folks, >> >> >> Glad to announce that the first version of Phux PHP Router is released. >> >> https://github.com/c9s/phux >> >> >> Phux is basically a rewrite of PHP Roller router. >> >> Phux provides full functional C extension and pure PHP library with the >> same interface, so the extension is optional to install. >> >> but if you install phux extension, you may enjoy the performance of >> routing. >> >> >> >> Route Dispatch Benchmark (Symfony/Routing and Phux pure PHP) >> >> https://raw.github.com/c9s/Phux/master/benchmarks/reqs.png > > > Hi, > > Not a comment about Phux directly, but I think I have to point this out > (sadly: once more): When you use Apache Benchmark to benchmark the > performance of an PHP-component, you in fact benchmark your system... [1] > Another flaw is, that the whole benchmark setup is missing, which makes it > slightly suspicious and unreliable. Two things, that spontanous came into > my mind > > - Because you use ab to benchmark it is unlikely, that the S2-setup > contains the routing-component only. This means, that in this benchmark you > measure more than just the routing. OK, the README states, that "Testing > with route dispatch only. (no controller)", but information on how do the > benchmarks setup the routing is missing. > - Like so many other benchmarks improvments, that every S2-application > relies in production on, are probably missing. Do you compile the > routing-table from a config into PHP? Do you use APC? While it sounds > totally fair to compare two bare setups, it isn't. It probably ignores > customizations, that are possible in one, but not the other and that > affects the outcome. Most obvious example is what I mentioned: Parsing a > YAML, generating the routing table and routing is something very different > than reading the routing-table from an APC-cached PHP-file and routing. > > I don't say Phux is bad or something, but my advice is to rework the > benchmark and publish the setup. Beside this I am more interested in > benchmarks of the C-extension (especially against the native-PHP solution) > :) > > Regards, > Sebastian > > > [1] The worst thing about this, is that you not only benchmark the target > system, but the benchmark also relies heavily on the local one. Even worse > than worse: Running both on the same machine let them affect each other in > a non-predictable way. You get numbers, but you can never tell, what they > mean. > > >> >> >> >> >> Cheers, >> >> Yo-An Lin >> > > > > -- > github.com/KingCrunch > > > -- github.com/KingCrunch
