Date: Wed, 24 Nov 2004 19:33:55 +0100 From: =?ISO-8859-1?Q?Eirik_=D8verby?= <ltning@anduin.net> To: Pawel Jakub Dawidek <pjd@FreeBSD.org> Cc: stable@freebsd.org Subject: Re: graid3 - requirements or manpage wrong? Message-ID: <6579E984-3E47-11D9-9576-000D9335BCEC@anduin.net> In-Reply-To: <20041124171115.GP7232@darkness.comp.waw.pl> References: <41A45A3F.5010008@anduin.net> <20041124171115.GP7232@darkness.comp.waw.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
On 24. Nov 2004, at 18:11, Pawel Jakub Dawidek wrote: > On Wed, Nov 24, 2004 at 10:54:07AM +0100, Eirik ?verby wrote: > +> to the best of my ability I have been investigating the 'real' > +> requirements of a raid-3 array, and cannot see that the following > text > +> from graid3(8) cannot possibly be correct - and if it is, then the > +> implementation must be wrong or incomplete (emphasis added): > +> > +> label Create a RAID3 device. The last given component will > contain > +> parity data, all the rest - regular data. ***Number > of > +> compo- > +> nents has to be equal to 3, 5, 9, 17, etc. (2^n + > 1).*** > +> > +> I might be wrong, but I cannot see how a raid-3 array should > require > +> (2^n + 1) drives - I am fairly certain I have seen raid-3 arrays > +> consisting of four drives, for example. This is also what I had > hoped to > +> accomplish. > > This requirement is because we want sectorsize to be power of 2 > (UFS needs it). > In RAID3 we want to send every I/O request to all components at once, > that's why we need sector size to be N*512, where N is a power of 2 > value > AND because graid3 uses one parity component we need N+1 providers. OK I see, makes sense. So it's not really a raid3 issue, but an implementation issue. The only problem then is - gvinum being in a completely unusable state (for raid5 anyway), what are my alternatives? I have four 160gb IDE drives, and I want capacity+redundancy. Performance is a non-issue, really. What do I do - in software? /Eirik > > > -- > Pawel Jakub Dawidek http://www.FreeBSD.org > pjd@FreeBSD.org http://garage.freebsd.pl > FreeBSD committer Am I Evil? Yes, I Am! On 24. Nov 2004, at 18:11, Pawel Jakub Dawidek wrote: On Wed, Nov 24, 2004 at 10:54:07AM +0100, Eirik ?verby wrote: +> to the best of my ability I have been investigating the 'real' +> requirements of a raid-3 array, and cannot see that the following text +> from graid3(8) cannot possibly be correct - and if it is, then the +> implementation must be wrong or incomplete (emphasis added): +> +> label Create a RAID3 device. The last given component will contain +> parity data, all the rest - regular data. ***Number of +> compo- +> nents has to be equal to 3, 5, 9, 17, etc. (2^n + 1).*** +> +> I might be wrong, but I cannot see how a raid-3 array should require +> (2^n + 1) drives - I am fairly certain I have seen raid-3 arrays +> consisting of four drives, for example. This is also what I had hoped to +> accomplish. This requirement is because we want sectorsize to be power of 2 (UFS needs it). In RAID3 we want to send every I/O request to all components at once, that's why we need sector size to be N*512, where N is a power of 2 value AND because graid3 uses one parity component we need N+1 providers. OK I see, makes sense. So it's not really a raid3 issue, but an implementation issue. The only problem then is - gvinum being in a completely unusable state (for raid5 anyway), what are my alternatives? I have four 160gb IDE drives, and I want capacity+redundancy. Performance is a non-issue, really. What do I do - in software? /Eirik -- Pawel Jakub Dawidek http://www.FreeBSD.org pjd@FreeBSD.org http://garage.freebsd.pl FreeBSD committer Am I Evil? Yes, I Am!
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6579E984-3E47-11D9-9576-000D9335BCEC>