Return to BSD News archive
#! rnews 2106 bsd From: Michael Hancock <michaelh@cet.co.jp> Newsgroups: comp.programming.threads,comp.unix.bsd.freebsd.misc Subject: Re: [??] pure kernel vs. dual concurrency implementations Date: Sun, 23 Feb 1997 14:39:44 +0900 Organization: CET Lines: 33 Message-ID: <330FD820.124A@cet.co.jp> References: <330CE6A4.63B0@cet.co.jp> <330E8716.167EB0E7@nyx.pr.mcs.net> NNTP-Posting-Host: a05m.cet.co.jp Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Mailer: Mozilla 2.0 (Win95; I) Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!ns.saard.net!spasun.tpa.com.au!duster.adelaide.on.net!news.ade.connect.com.au!news.mel.connect.com.au!munnari.OZ.AU!uunet!in1.uu.net!192.244.176.52!news.iij.ad.jp!news.CET.CO.JP!usenet Xref: euryale.cc.adfa.oz.au comp.programming.threads:3287 comp.unix.bsd.freebsd.misc:36023 Chris Csanady wrote: > > Michael Hancock wrote: > > > > I've been talking to some people who are pro pure kernel threading vs. > > a dual kernel and userland model when it comes to implementation on a > > traditional Unix kernel design like FreeBSD. > > > > There was a discussion on the hackers a while ago with the subject > "Re: clone()/rfork()/threads (Re: Inferno for FreeBSD)" that you > might want to look at. I didn't follow it closely, but Terry explained > a bunch of stuff, and mentioned that the DEC model is what we want.. I read that and DEC's two-level implementation sounds impressive and clean. However, for FreeBSD I think I have to agree with others that the 1 to 1 model is the way to go for the following reasons. 1) The simplicity will help get bugs out of the development cycle sooner. We still aren't using a reentrant libc and this and other areas will need time and effort. I rather not go through 2 to 3 years of instability. 2) We will get more quality improvements to the kernel as it is made more flexible to handle processes, threads, and in-betweenies. 3) The resident performance freaks will have more incentive to find ways to get context switch times even lower. Regards, Mike Hancock