httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Trawick <traw...@attglobal.net>
Subject Re: Tagged the tree
Date Thu, 22 May 2003 11:29:32 GMT
Jeff Trawick wrote:
> Jeff Trawick wrote:
> 
>> Sander Striker wrote:
>>
>>> I just tagged the tree with STRIKER_2_0_46_PRE1.  Please test.
> 
> 
> time to disagree with myself...  starting to see a performance 
> regression w.r.t. 2.0.45 on AIX with simple static file serving... 
> degradation is on the order of 2-3%

this is weird, and will very possibly turn out to be something stupid 
I'm doing, but it is so consistent

I started playing on RH 9, and also on that box 2.0.46-pre1 is 
consistently slower than 2.0.45 (though not as dramatically as on 
AIX)...  interestingly: plug new apr/aprutil into 2.0.45 and it gets a 
little faster, plug old apr/aprutil into 2.0.46-pre1 and it gets a 
little slower; so there is some good news out of this :)

2.0.45's ab has a storage leak seen over 100,000 requests... it ends up 
with more storage than an Apache worker child process after a while... 
presumably 2.0.46-pre1's ab does too

2.0.46-pre1 httpd doesn't seem to be leaking storage

looking at diffs between the trees...

the new get_server_name_for_url() has an extra strchr() if 
APR_HAVE_IPV6, but we don't get there for a straight static page served 
to ab

the disable nagle stuff seems to be right (disabled on the listening 
socket, calls to apr are noops on connected sockets, just as with 2.0.45)

time to learn oprofile I guess


Mime
View raw message