Date: Fri, 6 Jul 2012 07:52:00 -0700 (PDT) From: Jason Usher <jusher71@yahoo.com> To: Zaphod Beeblebrox <zbeeble@gmail.com> Cc: freebsd-fs@freebsd.org Subject: Re: vdev/pool math with combined raidzX vdevs... Message-ID: <1341586320.93244.YahooMailClassic@web122502.mail.ne1.yahoo.com> In-Reply-To: <CACpH0MemwZDCXsh4USzeFHUO8fbW09TSOYyVPa2dWmKc8N%2B=_Q@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
=0A=0A--- On Fri, 7/6/12, Zaphod Beeblebrox <zbeeble@gmail.com> wrote:=0A= =0A> Is there some penalty for not=0A> googling some basic stats course?=A0= OK.=0A> This is from memory (hint: you probably should google).=0A=0A=0AAc= tually I spent a few hours googling zfs probabilities, and I cannot find an= y discussion of the changes in failure probability when multiple raidzX are= striped in a single pool.=0A=0AI did not try to google the math and do it = myself, as I wouldn't trust my own results.=0A=0A=0A> Similarly,=0A> =0A> p= (12drz2) =3D 12 * p(f) * 11 * p(f) * 10 * p(f)=0A> p(12drz3) =3D 12 * p(f) = * 11 * p(f) * 10 * p(f) * 9 * p(f)=0A> =0A> ... again with those assumption= s are more complex=0A> probabilities given=0A> your replacement strategy.= =0A> =0A> ... so, again with simplistic assumptions,=0A> =0A> p(36drz3 --- = 12 drives, 3 groups) =3D p(12drz3) * 3=0A> =0A> A "vanilla" RAID-Z2 (if I m= ake an assumption to what you're=0A> saying) is:=0A> =0A> p(36drz2) =3D 36 = * p(f) * 35 * p(f)=0A> =0A> ... but I can't directly answer you question wi= thout knowing=0A> a) the=0A> structure of the RAID-Z2 array and p(f).=A0 If= we use a=0A> 1% figure for=0A> p(f), then P(36drz3,12,3) =3D 0.035% and p(= 36drz2) =3D 4.3%=0A> =0A> ... that is the raid-Z2 case (one group of 36 dri= ves, two=0A> redundant=0A> --- which is crazy) is 4.3% likely to fail where= the 3-group=0A> RAID-Z3=0A> is only 0.035% likely to fail.=A0 As a more sa= ne=0A> comparison,=0A> p(36drz2,12,3) =3D 3.8%=0A=0A=0AOk, you're right - I= did not specify the structure of the raidz2 array.=0A=0AWhat I meant to co= mpare was the failure probabilities of:=0A=0A- a single raidz2 vdev made up= of 12 disks (10 data, 2 parity)=0A=0Avs.=0A=0A- a single raidz3 vdev made = up of 12 disks (9 data, 3 parity)=0A=0Avs.=0A=0A- a single raidz3 vdev made= up of 12 disks (9 data, 3 parity) which ALSO happens to be participating i= n a stripe with two other identical raidz3 vdevs, all in one zpool.=0A=0A= =0AI can see some probabilites for the first two examples here:=0A=0Ahttp:/= /hardforum.com/showthread.php?t=3D1621123=0A=0A(search for first post on pa= ge by "john4200")=0A=0Aas well as your own math, which appears to be the sa= me.=0A=0A=0AAlso, I think we don't need to specify the failure rate, F, sin= ce we are merely comparing three scenarios, and can compare results that st= ill contain an 'F' variable in them ... that is, our answers can contain a = yet undefined 'F', right ?=0A=0A=0AAs for myself, I have decided that raidz= 2 is "not enough" for me, and at the same time, would really, really like t= o combine three raidz3 into a single zpool ... but I don't want to do that = if that configuration brings me back to raidz2-ish failure probabilities...
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1341586320.93244.YahooMailClassic>