www-apache-bugdb mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: os-irix/1660: accept(2) serialization patch using MIPS ABI; two orders of magnitude faster than System V semaphores.
Date Tue, 13 Jan 1998 06:30:00 GMT
The following reply was made to PR os-irix/1660; it has been noted by GNATS.

From: Dean Gaudet <dgaudet@arctic.org>
To: Paul Walmsley <ccshag@cclabs.missouri.edu>
Cc: apbugs@hyperreal.org
Subject: Re: os-irix/1660: accept(2) serialization patch using MIPS ABI; two orders of magnitude
faster than System V semaphores.
Date: Mon, 12 Jan 1998 22:38:10 -0800 (PST)

 Did you give -DUSE_USLOCK_SERIALIZED_ACCEPT a try?  I found it to be a
 huge win on multiprocessor IRIX boxes, but a huge loss on single processor
 boxes (allow me to curse SGI).  It's mentioned in the perf-tuning docs.
 
 Have you tested your code on both multiprocessor and uniprocessor boxes? 
 
 Another thing -- I'm guessing that the kernel does not clean up those mips
 abi locks if the task holding it exits.  If that's the case then you'll
 need to add code a lot like the pthreads mutex code has to do cleanup on
 task kill.
 
 You may want to grab a recent snapshot of the 1.3b4-dev code from
 ftp://dev.apache.org/from-cvs/ -- there's a file src/test/test-sem.c which
 is a harness for testing the various semaphores that mimics the
 fork()/multiple child behaviour of apache. 
 
 Dean
 
 

Mime
View raw message