Subject: Re: SOFTDEP, a second data point
To: Jim Wise <jwise@draga.com>
From: Nathan J. Williams <nathanw@MIT.EDU>
List: tech-kern
Date: 12/13/1999 15:53:59
<jwise@draga.com> (Jim Wise) writes:

> In addition, in order to get some performance numbers slightly more
> useful than `it feels faster', I ran the following test on my home
> fileserver / i386 build server:
> 
> 	1.) rebuild kernel with softdep compiled in.
> 	2.) make build
> 	3.) time /usr/bin/make build 2>&1 | tee build.1.log
> 	4.) reboot off of /altroot and enable softdep on all drives,
> 	    / included.
> 	5.) time /usr/bin/make build 2>&1 | tee build.2.log

> The results were as follows:
> 
> The first build (softdep not enabled anywhere):
> 	    11302.97 real      6225.54 user      1850.15 sys
> 
> The second build (softdep enabled on all filesystems, including / (which
> includes /tmp):
> 	     9309.22 real      6179.97 user      1935.87 sys
> 
> 11302.97 seconds (~ 3:08:23) was reduced to 9309.22 seconds (~ 2:35:09),
> a speed improvement of 18%.  Not too shabby, IMHO.


It would be interesting to see the time for a build with all
filesystems mounted async, since one of the claims for softdep is that
it comes very close to that theoretical peak performance.

        - Nathan