I agree that Firebug times are erratic. I have put much of this down to
normal network lag/server lag. I am running a windows server running PHP
using the DLL version, and I think it sometimes gets locked during the
request. I am looking into getting FastCGI working to see if it is more
consistant but have never been fond of recompiling and XAMPP doesn't seem to
like to load any of the versions out there.
Don't suppose anyone has any experience of FastCGI under windows? Is it
worth it over the DLL? Does it make the server more stable? I'm running PHP
5.2.0 with APC at the mo.
If the timings of loading the user style sheet is wrong, then fine. Might be
worth hacking in an optional timing into output=js as a JS comment, to prove
it once and for all.
On 1/31/07, Nick Jenkins <nickpj(a)gmail.com> wrote:
Well I'm not running squid, and here's the test script I'm running and the
timing results I'm seeing on my local test wiki:
--------------------------------------
root@bling:~/tmp# cat test.sh
#!/bin/bash
time wget --output-document=index-1.js
"http://192.168.0.64/wiki/index.php?title=-&action=raw&smaxage=0&gen=js
"
time wget --output-document=index-2.css
"http://192.168.0.64/wiki/index.php?title=MediaWiki:Common.css&usemsgcache=yes&action=raw&ctype=text/css&smaxage=18000
"
time wget --output-document=index-3.css
"http://192.168.0.64/wiki/index.php?title=-&action=raw&gen=css&maxage=18000&smaxage=0&ts=20070129214411
"
root@bling:~/tmp#
root@bling:~/tmp# ./test.sh
--12:40:44--
http://192.168.0.64/wiki/index.php?title=-&action=raw&smaxage=0&…
=> `index-1.js'
Connecting to 192.168.0.64:80 <http://192.168.0.64/>... connected.
HTTP request sent, awaiting response... 200 OK
Length: 3,788 (3.7K) [text/javascript]
12:40:45 (212.50 MB/s) - `index-1.js' saved [3788/3788]
real 0m0.237s
user 0m0.004s
sys 0m0.000s
--12:40:45--
http://192.168.0.64/wiki/index.php?title=MediaWiki:Common.css&usemsgcac…
=> `index-2.css'
Connecting to 192.168.0.64:80 <http://192.168.0.64/>... connected.
HTTP request sent, awaiting response... 200 OK
Length: 28 [text/css]
12:40:45 (2.97 MB/s) - `index-2.css' saved [28/28]
real 0m0.198s
user 0m0.008s
sys 0m0.004s
--12:40:45--
http://192.168.0.64/wiki/index.php?title=-&action=raw&gen=css&m…
=> `index-3.css'
Connecting to 192.168.0.64:80 <http://192.168.0.64/>... connected.
HTTP request sent, awaiting response... 200 OK
Length: 75 [text/css]
12:40:45 (8.94 MB/s) - `index- 3.css' saved [75/75]
real 0m0.228s
user 0m0.004s
sys 0m0.004s
root@bling:~/tmp#
--------------------------------------
I.e. total time = 0.237 + 0.198 + 0.228 = 0.663 seconds.
I take it for the Firebug timing values, that you're going Tools ->
Firebug -> Open Firebug -> click on "Net" -> make sure "All"
is
selected -> read the timing values for the various scripts?
If so, my times are 10 ms, 150 ms, and 160 ms, respectively for the above
3 files.
However, ajaxwatch.js for me shows as taking 351 ms, which seems rather
long.
However, when I try to fetch ajaxwatch on the command line, I get a very
different result:
--------------------------------------
root@bling:~/tmp# time wget --output-document=ajaxwatch.js
"http://192.168.0.64/wiki/skins/common/ajaxwatch.js
"
--12:56:20--
http://192.168.0.64/wiki/skins/common/ajaxwatch.js
=> `ajaxwatch.js'
Connecting to 192.168.0.64:80 <http://192.168.0.64/>... connected.
HTTP request sent, awaiting response... 200 OK
Length: 3,916 (3.8K) [application/x-javascript]
12:56:20 (219.68 MB/s) - `ajaxwatch.js' saved [3916/3916]
real 0m0.008s
user 0m0.004s
sys 0m0.004s
root@bling:~/tmp#
--------------------------------------
I.e. only 2.3% of the time that Firebug reports. A 43-times difference
like this is an enormous difference - the difference between
blazingly fast, and pig slow.
Refreshing the page shows that Firebug thinks Monobook.css took 791 ms;
however the command line repeatably shows values between
0.197s and 0.210s.
I think there may be a few problems or explanations here:
* The Firebug times I suspect include the time that Firefox spends
client-side processing the CSS or JS (i.e. not just the time to
fetch from the network). I.e. image rendering times, JavaScript execution
times, page layout times, etc. If that's correct, what
would be useful would be is if the Firebug times separated out network
fetch times, and local client processing times, for each
item.
* The Firebug times do not seem to be as repeatable as one would like for
systematic testing ( e.g. total page display times varied
between 2.5 seconds and 3.5 seconds). Part of this may be the Observer
effect (i.e. Firebug is embedded in Firefox, and the act of
observing Firefox and continuously reporting on its rendering times
probably makes Firefox slower, and/or introduces variability).
* I have gzip compression turned on, and probably Firefox uses this,
whereas wget probably does not.
Basically, I'm unsure what to make of the Firebug times, except to say
that the times Firebug reports do not immediately appear to
correlate well with the time taken to retrieve the raw content from the
network.
Also, on a related note, for requesting ajaxwatch.js, we should probably
be tacking on the "?52" version string thing somewhere
(which from the Firebug output it looks like we're not currently doing).
All the best,
Nick.
-----Original Message-----
From: wikitech-l-bounces(a)lists.wikimedia.org
[mailto:wikitech-l-bounces@lists.wikimedia.org]On Behalf Of Alex Powell
Sent: Wednesday, 31 January 2007 2:18 AM
To: Wikimedia developers
Subject: Re: [Wikitech-l] Performance of loading custom style sheets
Hi,
No not that. I have a 2 GHz core duo at both ends, runnning no AV
software
(!), or anything else. The machine is
"sufficient". Wiki pages without
those
two (stripped them out) take 180millis by
Firebug.
Testing against the wikipedia, it took 200millis on first hit, and then
16
millis on subsiqent. I suspect the squid caches
had it in their mits at
that
point. The page is pulled with some unusual
parameters, I was wondering
if
that was pushing the code down an unusual
path...
Alex
On 1/29/07, Ashar Voultoiz <hashar(a)altern.org> wrote:
>
> Alex Powell wrote:
> > Hi,
>
> > Mediawiki's performance
has improved substantially over the
versions.
> and
> > 1.9 seems the snappiest yet. When I upgraded I found I had some slow
> stuff
> > being loaded in addition to MW. Using Firebug beta (v cool BTW) I
> noticed
> > something odd. The index.php for the page loads blindingly fast, but
the
> > following:
>
> >
/index.php?title=-&action=raw&smaxage=0&gen=js<
>
>
> > /index.php?title=MediaWiki:
>
Common.css&usemsgcache=yes&action=raw&ctype=text/css&smaxage=18000<
>
http://www.exscien.com/internal/index.php?title=MediaWiki:Common.css&us…
>
> >
/index.php?title=User:Alexp/monobook.css&action=raw&ctype=text/css<
>
http://www.exscien.com/internal/index.php?title=User:Alexp/monobook.css&…
/index.php?title=-&action=raw&gen=css&maxage=18000&smaxage=0&ts=20070129214411<
>
http://www.exscien.com/internal/index.php?title=-&action=raw&gen=cs…
>
>
>
>
> > > ... seem to run very slowly (one was taking 2
seconds and each take
over
> > .5secs, on a page that took 16ms to
render). What can be done to
speed
> > them?
> > > As far as I can see they are just simple js direct from the DB. Am I
> > missing
> > > something or is there a performance hole there?
>
> > Get a better computer.
Seriously, I suspect firebug is timing the
> > firefox rendering time (for css) and execution time (for js). Try
using
> > wget, it will be faster (or disable javascript).
>
>
> > --
> > Ashar Voultoiz - WP++++
> >
http://en.wikipedia.org/wiki/User:Hashar
> >
http://www.livejournal.com/community/wikitech/
> > IM: hashar(a)jabber.org ICQ: 15325080
>
>
> > _______________________________________________
> > Wikitech-l mailing list
> > Wikitech-l(a)lists.wikimedia.org
> >
http://lists.wikimedia.org/mailman/listinfo/wikitech-l
>
>
_______________________________________________
> Wikitech-l mailing list
> Wikitech-l(a)lists.wikimedia.org
>
http://lists.wikimedia.org/mailman/listinfo/wikitech-l
_______________________________________________
Wikitech-l mailing list
Wikitech-l(a)lists.wikimedia.org
http://lists.wikimedia.org/mailman/listinfo/wikitech-l