httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: IE4 HTTP/1.1 Chunking bug reincarnation?
Date Tue, 27 Jan 1998 05:35:39 GMT


On Tue, 27 Jan 1998, Rasmus Lerdorf wrote:

> > It also includes another little performance tweak based on looking at
> > output from mod_php3.  If one of you fellows has the time, I'd love it if
> > you could play with various LARGE_WRITE_THRESHOLD settings to see what
> > value is good.  With the current apache cvs HEAD your module causes a
> > bunch of small chunks, like size 7, 0x1c, 0xa, just way small. 
> > 
> > With the 31 setting I put into this patch it generates output in nice ~4k
> > sized chunks. 
> 
> Well, it is probably only the show_source() code that generates those
> dinky little chunks.  That is the only place rputc() is really used.
> Normal script output does not do this and I am not at all concerned about
> performance of the show_source stuff.  That is mostly for debugging
> anyway, and it certainly has been handy for finding chunking problems! ;)

:)

Yeah but mod_include, mod_autoindex, and mod_status use rputc(), and
rputs() of small strings.  So it's a more general thing.  I was just
getting irked by seeing chunks of length 0xf57, then 0x19, then 0xffb,
then 0x7, and so on.  The writev() cost and the chunk creation cost just
don't seem worth it... the actual threshold is probably higher around
100... but 31 is a good start. 

Dean



Mime
View raw message