From owner-freebsd-current  Fri Jan 26 11:47:06 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id LAA21295
          for current-outgoing; Fri, 26 Jan 1996 11:47:06 -0800 (PST)
Received: (from dyson@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id LAA21289
          Fri, 26 Jan 1996 11:47:05 -0800 (PST)
From: John Dyson <dyson>
Message-Id: <199601261947.LAA21289@freefall.freebsd.org>
Subject: Re: Good news -- pipe stuff
To: terry@lambert.org (Terry Lambert)
Date: Fri, 26 Jan 1996 11:47:04 -0800 (PST)
Cc: current@freefall.freebsd.org
In-Reply-To: <199601261851.LAA04990@phaeton.artisoft.com> from "Terry Lambert" at Jan 26, 96 11:51:24 am
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-current@FreeBSD.ORG
Precedence: bulk

> 
> Good show on the pipe code!
> 
> I read the Lai/Baker paper: in "Figure 1", FreeBSD kicks some serious
> butt on context switches -- it appears to be both flat and linear past
> 200 processes (the limit of the graph in the figure).
> 
> I would be loathe to screw with what is essentially "perfection", IMO.
> Solaris is on the order of 3 times slower until 30 processes, and jumps
> to 6-9 times slower based on algorithm (the worst algorithm being the
> one they must use in the SMP case).
> 
I think that the context switch performance improvement that I would expect
to see would be mostly an artifact of the measurement method (improving
pipe perf.)  There should be NO negative impact, because all changes
will be in the pipe code (new file: sys_pipe.c) and removal of the
pipe() subroutine in uipc_syscalls.c.

John
dyson@freebsd.org