From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?ISO-8859-1?Q?Xu=E2n_Baldauf?= Subject: Re: Where to put the journal Date: Thu, 05 Aug 2004 19:05:21 +0200 Message-ID: <411268D1.7000101@baldauf.org> References: <410CCFCA.1030309@baldauf.org> <410E08F7.9020303@namesys.com> <411262DC.4070206@namesys.com> Mime-Version: 1.0 Content-Transfer-Encoding: quoted-printable Return-path: list-help: list-unsubscribe: list-post: Errors-To: flx@namesys.com In-Reply-To: <411262DC.4070206@namesys.com> List-Id: Content-Type: text/plain; charset="iso-8859-1"; format="flowed" To: Edward Shishkin Cc: "Vladimir V. Saveliev" , Reiserfs List Edward Shishkin wrote: > Vladimir V. Saveliev wrote: > >> Hello >> >> Xu=E2n Baldauf wrote: >> >>> Hello, >>> >>> I want to setup a Linux software RAID 5 over disks of different=20 >>> speed with reiserfs on top. I know that the journal has to be=20 >>> accessed frequently for comparably small accesses. That's why I'm=20 >>> considering to "offload" the journal onto an extra partition on a=20 >>> faster disk of the RAID set. >>> >>> Thus, the journal would not be protected by the RAID, but this is=20 >>> not so important for me. >>> >>> Is this kind of reasoning sane? Would there be any speedup by=20 >>> offloading the journal? >>> >> > > Hello. > Unfortunately comprehensive results of journal relocation are obsolete > as it was done at the beginning of mongo evolution. As I remember there > were no any essential speedups. The only result of journal =20 > relocation that > was created by new mongo benchmark can be found here: > http://thebsh.namesys.com/benchmarks/journal_relocation_to_NVRAM.html Thank you, essentially, these numbers say that relocation provided a speedup of=20 about 10% for write. This surprises me a bit, because I expected a=20 significantly smaller seek count for a write operation. Nevertheless,=20 maybe I should run benchmarks with many small files for myself. :-) > > Edward. ciao, Xu=E2n Baldauf. > >> Person who did some tests on non standard journal is on vacation. He=20 >> is expected to come back today. Maybe he remembers benchmark results, >