Hi, just retaking the conversion.. some feedback:
i found a problem when processing a high load of PHP scripts with a PHP-FPM backend, this is something to be addressed on Monkey side. About the benchmark that you are doing, i think is honest to say that at this point the FastCGI Monkey implementation have a bug to be addressed and until is not fixed should not be used to serve PHP on high end loads, that is fair and there is no problems to make that public. I assume PHP is not the only test that you are doing, there should be static files (small, medium big) and other things... so count us IN. This is an open project and every success and failure at any point should *not* be hidden, we have grown from every failure on the past so we are ready for the challenge. If there is something not working we will fix it, if things works great we will even keep improving. NOTE: when testing a short PHP response, make sure to DISABLE caching on every server to make it fair. regards, On Mon, Dec 30, 2013 at 1:39 PM, Eduardo Silva <[email protected]> wrote: > Hi, > > > On Mon, Dec 30, 2013 at 1:27 PM, Fnux <[email protected]> wrote: > >> Hi Eduardo, >> >> I didn't come back to you earlier since I wanted to be sure of my >> testings. >> >> As previously said, I'm doing a fair and independent comparative >> benchmark of different web servers and web application servers, including >> Monkey. >> >> I also said that I won't use you test tool since it's obviously one of >> the slowest I've ever seen, and even when using it, it doesn't give the >> best results for Monkey. >> >> The main reason of the poor performance of it is that it doesn't use the >> multi-thread capabilities of “modern” multi-core CPUs (and I should stop to >> say modern since multi-core CPUs are available now for a full decade). >> >> Instead, I use both weighttp (from Lighttpd) as well as abc, that's >> nothing but an open source weighttp wrapper developed the G-WAN team and >> allowing to run weighttp several times within a loop that increases the >> number of simultaneous clients. >> >> > since then some work have been done in our benchmark wrapper, the new one > can be found on: > > > http://git.monkey-project.com/?p=monkey-bench > > > >> Further, abc allows to save the benchmark results in CVS type files that >> you can then use to create graphics. >> >> > thats a pending missing feature to be included, just a FYI: as part as our > QA and CI tests, we are integrating a benchmark suite, this is a complete > project going on... > > > >> So, let's go back to your “Baby”. >> >> I'm very frustrated when trying to benchmark Monkey and any simple PHP >> application such the trivial "Hello World!" since Monkey stops to respond >> as soon as there is very few connections. >> >> I guess it's a FASTCGI problem that you should take a look at it urgently >> (even using only weighttp) since due to the lack of response of Monkey, >> your product will be put out of the race in my paper, and with the >> explanation why it's discarded. >> >> Since I don't want to hurt nor your team or you when publishing the >> reason of such a failure to compete even with the good old dog Apache2, I >> would appreciate your view point on this very specific subject. >> >> > Besides PHP, are you running other tests ?, if so why not to avoid just > the PHP ? > > Now, if you want to perform real, reliable, not biased and reproducible >> stress benchmarks of Monkey (as well as of any other web server), please >> use abc on your own test systems and then share with me what are your >> comparative results. >> >> They may differ from mine, but unfortunately I guess and bet not that >> much! >> >> Happy holiday season to both of your development team members, your >> friends and family and you. >> >> > thanks for your best wishes!, i share the same ones for you and family! > > > >> Note: To install abc, please do the following (this is valid on any >> Debian like distro as well as Ubuntu and its derivative such as LinuxMint) >> >> sudo apt-get -y build-dep libev >> sudo apt-get -y install libev-dev >> mkdir abc >> cd abc >> wget http://dist.chmorph.de/libev/libev-4.15.tar.gz >> tar -xzf libev-4.15.tar.gz >> ./configure >> make >> sudo make install >> wget http://as2.com/linux/tools/weighttp64.tar.bz2 >> tar -xjf weighttp64.tar.bz2 >> sudo cp weighttp64 /usr/local/bin/weighttp >> wget http://as2.com/linux/tools/abc.tar.bz2 >> tar -xjf abc.tar.bz2 >> gcc -02 abc.c -lpthread -o abc >> sudo cp abc /usr/local/bin >> >> Then you're done. >> >> > i will give it a try and see what is going on, i have not seen a similar > issue, before that i have a few questions: > > 1) which Monkey version are you using ? (1.3.0 or git ?) > > 2) what are you using to run PHP on the backend ?, are you using php-fpm > or any similar interface ? > > 3) for the backend that runs PHP, please provide the configuration file. > > 4) what is the specific commands for 'abc' that you use to test the > servers ? > > with that information i will be able to dig more into the issue, > > best, > > > > -- Eduardo Silva http://edsiper.linuxchile.cl http://monkey-project.com
_______________________________________________ Monkey mailing list [email protected] http://lists.monkey-project.com/listinfo/monkey
