*BSD News Article 18702


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!agate!toe.CS.Berkeley.EDU!bostic
From: bostic@toe.CS.Berkeley.EDU (Keith Bostic)
Newsgroups: comp.unix.bsd
Subject: Re: 4.4BSD - LFS performance ?
Date: 21 Jul 1993 12:15:43 GMT
Organization: University of California, Berkeley
Lines: 31
Message-ID: <22jc1f$nsm@agate.berkeley.edu>
References: <avalon.743239015@cairo>
NNTP-Posting-Host: toe.cs.berkeley.edu

In article <avalon.743239015@cairo> avalon@cairo.anu.edu.au (Darren Reed) writes:
>
>Now that 4.4BSD is released, are there any figures on what sort of
>performance difference between LFS and other FS's, especially FFS ?
>

An interesting question.  I'd strongly recommend reading Dr.
Margo Seltzer's doctoral thesis; it's up for anonymous ftp on
toe.cs.berkeley.edu:pub/personal/margo/thesis.ps.Z, or, you can
order copies from the University of California, Berkeley, CS
department.  It has some detailed analysis of file systems in
general, and the 4.4BSD LFS/FFS in particular.

My three-paragraph summary of adding LFS to 4.4BSD is that LFS,
when garbage collection is not running, is a faster file system
than the FFS provided with 4.3BSD.  However, adding Larry McVoy's
clustering changes to FFS (~500 lines of C) makes FFS perform
comparably to LFS.  When the LFS cleaner is running and garbage
collection is being done, LFS has large perturbations in its
performance characteristics, and is generally slower than FFS.

The future work is to smooth out the cleaner's effect on the
system.   (Or, sites can simply provide enough disk space so
that garbage collection becomes an idle-loop process.)

I still believe that LFS is worth doing, but because of its new
features (fast startup, version recovery, transactions, etc.),
not because of its performance characteristics.

--keith