Date: Tue, 01 Dec 2009 17:40:44 +0200 From: Alexander Motin <mav@FreeBSD.org> To: Johan Hendriks <Johan@double-l.nl> Cc: freebsd-stable@FreeBSD.org Subject: Re: Phoronix Benchmarks: Waht's wrong with FreeBSD 8.0? Message-ID: <4B1538FC.9090003@FreeBSD.org> In-Reply-To: <57200BF94E69E54880C9BB1AF714BBCBA572A3@w2003s01.double-l.local> References: <1259583785.00188644.1259572202@10.7.7.3> <4B153341.3060909@FreeBSD.org> <57200BF94E69E54880C9BB1AF714BBCBA572A3@w2003s01.double-l.local>
next in thread | previous in thread | raw e-mail | index | archive | help
Johan Hendriks wrote: >> If somebody still have questions, after some UFS parameters tuning I've >> got with the same tiotest tool: >> - Random Write latency - 15us, >> - Random Read latency - 7us. > > What kind of UFS parameter tunings. To maximize write-back delay. I've mounted file system asynchronously and increased vfs.dirtybufthresh and vfs.hidirtybuffers values ten times. > If things ar tuned for old hardware, which hardware are we talking about > i386? Or i486? > Maybe we should set the defaults for AMD64 in a way that modern hardware > can handle. > > AMD64 is a for modern hardware, it does not run on a pentium3. It is not a tuning for hardware. It is mostly tuning for sanity. Enormous write-back without using additional technics increase chance of data loss on power-outage, but doesn't give principal effect under constant load. This test is so easy to cheat, as it runs only few seconds and completely ignores cache effects. I believe that the same test with work file size increased by 100 times would show completely different results. -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4B1538FC.9090003>