Date: Sat, 14 May 2005 18:00:43 +0400 From: Vladimir Dzhivsanoff <vohand@gmail.com> To: Pawel Jakub Dawidek <pjd@freebsd.org> Cc: freebsd-stable@freebsd.org Subject: Re: gmirror Message-ID: <4200469905051407007484832@mail.gmail.com> In-Reply-To: <20050514131648.GB837@darkness.comp.waw.pl> References: <20050514093217.C6088E082A@oak.tantieme.ru> <20050514131648.GB837@darkness.comp.waw.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
On 5/14/05, Pawel Jakub Dawidek <pjd@freebsd.org> wrote: > On Sat, May 14, 2005 at 01:22:05PM +0400, vohand@gmail.com wrote: > +>=20 > +> Hardware: SATA RAID adapter with SiliconImage 3114 chip. 2 SATA HDD. > +> I did gmirror. > +>=20 > +> Run test: > +> 3 parallel tasks "dd if=3D/dev/mirror/gm0 of=3D/dev/null bs=3D1m count= =3D5000" > +>=20 > +> gstat shows summary read speed from mirror about 30-40 mbyte/sec. > +> It is for all balance algorithms: load, round-robin, split > +>=20 > +> While, one "dd if=3D/dev/ad4 of=3D/dev/null bs=3D1m count=3D5000" show= s 50 > mbyte/sec > +>=20 > +> 1. Why it occurs and how fix it ? > +>=20 > +> 2. If in the test instead of three instancies of "dd if=3D/dev/mirror/= gm0 > +> of=3D/dev/null bs=3D1m count=3D5000" run only one,=20 > +> then read speed don`t exceeds read speed from one disk (dd > if=3D/dev/ad4 > +> of=3D/dev/null bs=3D1m count=3D5000) > +> =09a) This is gmirror feature ? > +> =09b) This is hardware feature (SiliconImage 3114 chip) ? >=20 > This is the way RAID1 works. Try to imagine how disk's heads are moving - > there will be no speed-up in sequential reads, this is not RAID0. >=20 > Mirror characteristics are: > - the same speed for sequential reads as for one disk; > - the same speed for sequential/random write as for one disk; =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D > - double speed of one disk for random reads; by what test-suite I can test it ? and what balance algorithm is more proper for random reads ? three parallel tasks of "dd ...." is not good model for random reads ? >=20 > --=20 > Pawel Jakub Dawidek http://www.wheel.pl > pjd@FreeBSD.org http://www.FreeBSD.org > FreeBSD committer Am I Evil? Yes, I Am! >=20 >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4200469905051407007484832>