[mapserver-dev] More Performance

Paul Ramsey pramsey at cleverelephant.ca
Mon May 12 17:56:16 EDT 2008

Last post: under FCGI FT_New_Face disappears as a source of delay,
presumably because it's a one-time hit. So (of course) does all the
setup time for dylib on the system side. What's left is pretty hard to
squeeze more. Largely in the rendering library and image output
(compressing output).

Also, looking at the "Chart" view of the CPU usage during the sample
period (Shark is so cool!!!!) it appears that a render cycle is only
taking between 10ms and 15ms.

Great work, speed racers!


On Mon, May 12, 2008 at 1:02 PM, Paul Ramsey <pramsey at cleverelephant.ca> wrote:
> It occurs to me that this is a better screenshot. The 7.1% used by
> FT_New_Face happens with both GD and AGG. It's worth about 3-5ms, on
> my machine.
> P.
> On Mon, May 12, 2008 at 12:59 PM, Paul Ramsey <pramsey at cleverelephant.ca> wrote:
>> On Mon, May 12, 2008 at 12:32 PM, thomas bonfort
>> <thomas.bonfort at gmail.com> wrote:
>>> From the graph it seems like gd's font rendering/cache is a good
>>> culprit. Paul, is it easy for you to try the same thing with the agg
>>> backend
>> Since voodoo magic occurred and AGG just now compiled for me against
>> the MacPorts AGG, I can now report that: no, it doesn't make a
>> difference. The pain is occurring below AGG and GD, in the same place,
>> FT_New_Face, and of course AGG makes things slower in all kinds of
>> other ways, from having a 32-bit output to compress to spending more
>> time on the graphics drawing itself.
>> This is the latest 'svn update', so I can also report that danmo's
>> move of the curl init has had the desired effect.
>> P.

More information about the mapserver-dev mailing list