*BSD News Article 86262


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!feed1.news.erols.com!cpk-news-hub1.bbnplanet.com!news.bbnplanet.com!su-news-hub1.bbnplanet.com!venus.sun.com!uk-usenet.uk.sun.com!usenet
From: Casper.Dik@Holland.Sun.COM (Casper H.S. Dik - Network Security Engineer)
Newsgroups: comp.unix.bsd.misc,comp.unix.bsd.bsdi.misc,comp.unix.bsd.freebsd.misc,comp.unix.bsd.netbsd.misc
Subject: Re: Differences between BSDI, FreeBSD and NetBSD?
Supersedes: <cancel.casper.852470946@uk-usenet.uk.sun.com>
Date: 5 Jan 1997 13:32:15 GMT
Organization: Sun Microsystems, Netherlands
Lines: 33
Message-ID: <casper.852470946@uk-usenet.uk.sun.com>
References: <6OBgx1wrNgB@me-tech.PFM-Mainz.de> 	<1997Jan4.095835.23223@wavehh.hanse.de> <x7dhgkwius7.fsf@the-light-fantastic.MIT.EDU>
NNTP-Posting-Host: room101.holland.sun.com
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.misc:1879 comp.unix.bsd.bsdi.misc:5522 comp.unix.bsd.freebsd.misc:33654 comp.unix.bsd.netbsd.misc:5075

ghudson@mit.edu (Greg Hudson) writes:

>It's worth noting that historically most operating systems initially
>become much slower, for normal applications, when they integrate the
>VM and buffer cache.  Typically what happens is that one side of the
>cache starves the other for pages, and some frequent operation becomes
>very slow.  Proper tuning can solve the problem, but don't assume that
>an integrated buffer cache necessarily means improved performance; it
>just means more flexibility for the system at runtime.

Indeed.  E.g, originally in SunOS 4.0 and early Solaris versions, large
writes would consume all available pages stealing all non dirty pages
and forcing dirty pages to be flushed, causing all applications to suffer.
This was later remedied by allowing a single file to have at most N bytes
of outstanding writes.  That, unfortunately, has a drawback when you
*want* all your many MBs to work as a write ahead cache, so tuning
is *required* when you want to cache large write aheads.

(In Solaris 2.x, you can both increase the size which at 384K is pretty small,
IMHO, and switch off this throttling behaviour entirely)

The flexibility is important; earlier the kernel would simply steal a
percentage of memory for file buffering, whether you needed that much, more
or much less.  With a unified buffer/page space you have all memory
available for all tasks, but since programs can produce much more data
typically than disks can consume, you need to be careful.

Casper
--
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.