[pLog-svn] resserver usage

Oscar Renalias oscar at renalias.net
Thu Oct 5 12:58:21 GMT 2006


One thing I didn't measure in my tests was the CPU load when using
fopen()+fread() or direct access to the file via the web server. Maybe
that's the reason? resserver.php has to streams big files while the
rest of the blog is serving small-ish pages (20-30kb compared to
300-400kb sometimes?)

On 10/5/06, Jon Daley <plogworld at jon.limedaley.com> wrote:
>         But if it is the same, why is the load so much higher with the
> resserver?
>
> On Thu, 5 Oct 2006, Oscar Renalias wrote:
> > Yes we do it in exactly the same way, but at least in those cases I
> > don't see a clear way to improve it. At least with resources we're
> > considering direct access which would help alleviate the situation.
> >
> > On 10/5/06, Jon Daley <plogworld at jon.limedaley.com> wrote:
> >>         But, don't we do that sort of stuff on other requests?  ie. when I
> >> use custom URLs, I don't have the blogId, etc.  Maybe there needs to be
> >> some different/new indexes?
> >>
> >> On Thu, 5 Oct 2006, Oscar Renalias wrote:
> >>
> >> > I may try with fpassthru later on, but I am not sure whether it will
> >> > make any difference. In the tests I run I noticed that in the context
> >> > of 100 requests, usign fread() was only 1 second slower than using
> >> > direct URL access so this probably means that the problem is
> >> > elsewhere.
> >> >
> >> > Right now resserver.php needs to load a bunch of classes, validate
> >> > parameters, execute the pipeline for security stuff, load the plugin
> >> > manager and throw two events, load the blog id based on the name if
> >> > we're using subdomains, parse nice album names, etc. As soon as I
> >> > removed some of that sutff, we managed to save 15 seconds in 100
> >> > requisitions and over 1 minute when using altresserver.php, which only
> >> > implements the best case scenario: no subdomains and the blog id and
> >> > the resource name are in the request (no need to load any extra
> >> > classes and parse anything)
> >> >
> >> > What I am trying to say is that the bottleneck isn't the file-related
> >> > function calls but all what we have to do before we even get there.
> >> >
> >> > On 10/5/06, Jon Daley <plogworld at jon.limedaley.com> wrote:
> >> >>         You might want to include fpassthru in your tests - that is
> >> >> supposed to be the "recommended" way of doing this sort of thing.  I
> >> >> couldn't tell any difference when I changed the current resserver to use
> >> >> it though.
> >> >>
> >> >>
> >> >> On Thu, 5 Oct 2006, Oscar Renalias wrote:
> >> >> > Some more benchmarks I was working on today, tested with curl loading
> >> the
> >> >> > same URL 100 times (a 450kb JPEG file)
> >> >> >
> >> >> > testfopen.php
> >> >> > -------------
> >> >> > real    0m11.127s
> >> >> > user    0m1.205s
> >> >> > sys     0m1.893s
> >> >> >
> >> >> > direct url
> >> >> > ----------
> >> >> > real    0m10.170s
> >> >> > user    0m1.171s
> >> >> > sys     0m1.846s
> >> >> >
> >> >> > resserver.php 1.1.1
> >> >> > -------------------
> >> >> > real    1m39.753s
> >> >> > user    0m1.209s
> >> >> > sys     0m2.000s
> >> >> >
> >> >> > optimized resserver.php 1.1.1
> >> >> > -----------------------------
> >> >> > real    1m24.475s
> >> >> > user    0m1.210s
> >> >> > sys     0m2.007s
> >> >> >
> >> >> > altresserver.php
> >> >> > ----------------
> >> >> > real    0m36.123s
> >> >> > user    0m1.230s
> >> >> > sys     0m2.012s
> >> >> >
> >> >> > testfopen.php is a test to see how fast a raw fopen()+fread() is in
> >> PHP,
> >> >> and
> >> >> > it's a useless version of resserver.php. "Direct url" downloads the
> >> URL
> >> >> via
> >> >> > the web server, "resserver.php" is the default version of
> >> resserver.php
> >> >> > included in 1.1.1, "optimized resserver.php" very similar to the code
> >> in
> >> >> > 1.1.1 but removed a few things that could potentially be removed, and
> >> >> finally
> >> >> > "altresserver.php" is a very small version of resserver.php that still
> >> >> loads
> >> >> > a few classes (like GalleryResources and GalleryResource)
> >> >> >
> >> >> > Not much to comment on these figures, other than we're currently using
> >> >> the
> >> >> > slowest method and that direct url would still be the preferred way.
> >> >> >
> >> >> > At this point I am not sure if it's a good idea to create complex
> >> rewrite
> >> >> > rules or use obscure modules that no user will probably ever have
> >> >> available,
> >> >> > and instead we should concentrate on doing it properly once and for
> >> all.
> >> >> Some
> >> >> > of these workarounds could be used as temporary solutions while we fix
> >> >> what
> >> >> > is fundamentally slow (or broken), and that is the way in which we
> >> handle
> >> >> and
> >> >> > serve files.
> >> >> >
> >> >> > By the way I've attached the files I used for these tests, although if
> >> >> you
> >> >> > want to run them you'll need to make a few changes (file paths, URLs,
> >> >> etc)
> >> >> >
> >> >>
> >> >> --
> >> >> Jon Daley
> >> >> http://jon.limedaley.com/
> >> >>
> >> >> Teachers open the door, but you must enter by yourself.
> >> >> -- Chinese Proverb
> >> >> _______________________________________________
> >> >> pLog-svn mailing list
> >> >> pLog-svn at devel.lifetype.net
> >> >> http://devel.lifetype.net/mailman/listinfo/plog-svn
> >> >>
> >> > _______________________________________________
> >> > pLog-svn mailing list
> >> > pLog-svn at devel.lifetype.net
> >> > http://devel.lifetype.net/mailman/listinfo/plog-svn
> >> >
> >>
> >> --
> >> Jon Daley
> >> http://jon.limedaley.com/
> >>
> >> Research is what I'm doing when I don't know what I'm doing.
> >> -- von Braun
> >> _______________________________________________
> >> pLog-svn mailing list
> >> pLog-svn at devel.lifetype.net
> >> http://devel.lifetype.net/mailman/listinfo/plog-svn
> >>
> > _______________________________________________
> > pLog-svn mailing list
> > pLog-svn at devel.lifetype.net
> > http://devel.lifetype.net/mailman/listinfo/plog-svn
> >
>
> --
> Jon Daley
> http://jon.limedaley.com/
>
> If you have any trouble sounding condescending,
> find a Unix user to show you how it's done
> -- Scott Adams
> _______________________________________________
> pLog-svn mailing list
> pLog-svn at devel.lifetype.net
> http://devel.lifetype.net/mailman/listinfo/plog-svn
>


More information about the pLog-svn mailing list