FreeBSD shines..[Fwd: Re: semaphore speed] (fwd)

[Marc Slemko <ma...@znep.com>]

---------- Forwarded message ----------
Date: Mon, 03 Nov 1997 19:33:33 -0800
From: Julian Elischer <ju...@whistle.com>
To: hackers@FreeBSD.ORG
Subject: FreeBSD shines..[Fwd: Re: semaphore speed]

oops I sent this to the wrong address before..

participants:
jallison: Samba guru:
Andrew Tridgell: wrote Sambe.. Linux hacker
Julian: yours truely:
----------------results of benchmark--------------
/*

Results with the following defines:
  #define NSEMS 20
  #define NPROCS 60
  #define NUMOPS 10000

OSF1 dominion.aquasoft.com.au V4.0 564 alpha (233 MHz)
fcntl: 132.814 secs
ipc: 13.5186 secs

Linux 2.1.57 on a P120
fcntl: 21.3006 secs
ipc: 93.9982 secs

Solaris 2.5 on a Ultra170
fcntl: 192.805 secs
ipc: 18.6859 secs

IRIX 6.4 on a Origin200
fcntl: 183.488 secs
ipc: 10.4431 secs

SunOS 4.1.3 on a 2 CPU sparc 10
fcntl: 198.239 secs
ipc: 150.026 secs

Solaris 2.5.1 on a 4 CPU SPARCsystem-600
fcntl: 312.025 secs
ipc: 24.5416 secs

FreeBSD 2.2.2 on a 200MHz pentium
fcntl: 3.16579 secs
ipc: 2.63504 secs

Linux 2.0.30 on a PPro200
fcntl: 12.2177 secs
ipc: 50.4559 secs

 */

-------- andrew's resonse to included mail (see at end)-------
 (gee I hate how mime is handled in my mailer)

> so I 'spose now you'll have to go fix it so that we can't bask in
> our little glorie ;-)

Well, only if I have a spare afternoon, which isn't likely anytime soon.

The IPC code in Linux is really ugly. I had to hack it a bit to get
Oracle working under SparcLinux. I won't look at it again unless I
have to.

The real surprise is the fcntl locking speed. You either have some
really neat data structures or you don't do all the deadlock checking
linux does.
 
> of course I've never used them so I can't Guarantee that they are
> actually FUNCTIONNING correctly under FreeBSD :)

yeah, it's always easy to be fast when it doesn't have to be correct :-)

If anyone wants the code (and results for other OSes) see
ftp://samba.anu.edu.au/pub/tridge/semspeed.c

Cheers, Andrew

------------------ earlier mail------------------