Return to BSD News archive
Xref: sserve comp.unix.admin:14340 comp.unix.bsd:12775 comp.unix.internals:6357 comp.unix.wizards:31273
Newsgroups: comp.unix.admin,comp.unix.bsd,comp.unix.internals,comp.unix.wizards
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!elroy.jpl.nasa.gov!sdd.hp.com!nigel.msen.com!caen!batcomputer!cornell!chrisb
From: chrisb@cs.cornell.edu (Chris Buckley)
Subject: Re: why is altering MINFREE discouraged ?
Message-ID: <1993Oct20.143941.7910@cs.cornell.edu>
Organization: Cornell Univ. CS Dept, Ithaca NY 14853
References: <1993Oct15.212623.14509@aw101.iasl.ca.boeing.com> <PCG.93Oct18191520@decb.aber.ac.uk> <CF5DG9.6F@festival.ed.ac.uk> <PCG.93Oct19194620@decb.aber.ac.uk>
Date: Wed, 20 Oct 1993 14:39:41 GMT
Lines: 31
pcg@aber.ac.uk (Piercarlo Grandi) writes:
>>>> On Tue, 19 Oct 1993 13:57:41 GMT, richard@castle.ed.ac.uk (Richard
>>>> Tobin) said:
>Richard> In article <PCG.93Oct18191520@decb.aber.ac.uk> pcg@aber.ac.uk
>Richard> (Piercarlo Grandi) writes:
>pcg> Briefly: when looking for a cylinder group with free blocks to
>pcg> extend a file, the FFS chooses a cylinder group *at random*, for
>pcg> some rather good reasons. If the amount of free space is below a
>pcg> certain threshold, it must try several cylinder groups to find one
>pcg> with free blocks, and this is rather expensive.
>Richard> Just how expensive is it?
>Very expensive -- both in CPU time and reduced thruput. The FFS paper
>shows that it can halve the latter and substantially raise the former.
As always, I suspect it's very application dependent. When
reading/writing large sequential files there's no reason not to
use 1% free instead of 10% (or so my tests indicate). I can
easily believe that on a more typical many-user disk the
percentage should be higher; though previous times that this
subject has come up nobody has reported problems using 3-5% on
modern disks.
ChrisB
--
Chris Buckley Dept of Computer Science
chrisb@cs.cornell.edu Upson Hall, Cornell University
(609) 275-4691 Ithaca, NY 14852