Date: Thu, 17 Sep 1998 18:06:42 +0200
From: jum@helios.de (Jens-Uwe Mager)
Message-Id: <199809171606.SAA02095@isis.helios.de>
To: new-httpd@apache.org
Subject: Re: PR#2829 and related ones (semctl)
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org


>Yeah like two orders of magnitude better if the IRIX box has multiple
>CPUs.  But that was ages ago.  Things may have changed.  And it's clear
>that IRIX isn't set up by default to support sysvsem.  So changing it back
>to whatever was used in 1.2 seems fine... there's a bunch of docs to go
>hit though (upgrading_to_1.3, new_features_1.3, perf_tuning... maybe
>more).

I have had good success on our software by using abilock mechanism, it
is available on IRIX 5 and 6. From man abilock:

NAME
     init_lock, acquire_lock, release_lock, stat_lock, spin_lock - ABI mutual
     exclusion primitives

SYNOPSIS
     #include <abi_mutex.h>
 
     int init_lock(abilock_t *lck);
 
     int acquire_lock(abilock_t *lck);
 
     int release_lock(abilock_t *lck);
 
     int stat_lock(abilock_t *lck);
 
     void spin_lock(abilock_t *lck);


These do not use kernel calls, just atomic instructions of the
underlying processor (MIPS). I did an sginap(0) to reschedule on
contention while polling.


Jens-Uwe Mager

HELIOS Software GmbH
Steinriede 3
30827 Garbsen
Germany

Phone:		+49 5131 709320
FAX:		+49 5131 709325
Internet:	jum@helios.de