Date: Thu, 11 Sep 1997 01:33:35 -0700 (PDT) From: Dmitry Kohmanyuk <dk@dog.farm.org> To: spork@super-g.com (spork) Cc: freebsd-hackers@freebsd.org Subject: Re: mmap and INN Message-ID: <199709110833.BAA06048@dog.farm.org>
next in thread | raw e-mail | index | archive | help
In article <Pine.BSF.3.96.970911013258.16743A-100000@super-g.inch.com> you wrote: > I've attached two messages I dug up out of the mail archives that seem to > contradict each other. Gary says mmap + INN is a good thing, and Mike > says it's a bad thing... > Who should I listen to? We're close to going live with a 2.2-stable news > machine, and I'm a bit confused about what INN optimizations to go with. I am running FreeBSD 2.2.1-RELEASE with INN 1.5.1 and mmap (default setup from ports, only paths changed), and I indeed get zero-filled 4K blocks in my active file. So my advice is that mmap() is broken. I also have this suggestion: From: Ollivier Robert <roberto@keltia.freenix.fr> Message-ID: <19970720135853.17655@keltia.freenix.fr> <<< > hmm... why is uses > ## Should we msync when using mmap? Pick DO or DONT. Useful > ## with some slightly broken mmap implementations. (like HPUX and BSD/OS). > #### =()<MMAP_SYNC @<MMAP_SYNC>@>()= > MMAP_SYNC DONT > ?? > > should we declare FreeBSD mmap slightly broken?? ;-) Yes. You must use this in addition of the MSYNC_3_ARG parameter. That doesn't prevent the problem from happening though. >>> I haven't tried it yet; the problem with active file doesn't manifest itself often enough. I plan to upgrade to 1.6 and see if the problem persists, and then try MMAP_SYNC DONT.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709110833.BAA06048>