Date: Thu, 8 Nov 2012 19:57:11 -0600 From: "Matthew D. Fuller" <fullermd@over-yonder.net> To: Tom Evans <tevans.uk@googlemail.com> Cc: FreeBSD FS <freebsd-fs@freebsd.org> Subject: Re: SSD recommendations for ZFS cache/log Message-ID: <20121109015711.GB66994@over-yonder.net> In-Reply-To: <CAFHbX1K-NPuAy5tW0N8=sJD=CU0Q1Pm3ZDkVkE%2BdjpCsD1U8_Q@mail.gmail.com> References: <CAFHbX1K-NPuAy5tW0N8=sJD=CU0Q1Pm3ZDkVkE%2BdjpCsD1U8_Q@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Nov 08, 2012 at 09:07:24PM +0000 I heard the voice of Tom Evans, and lo! it spake thus: > > I've read that the Sandforce based devices get their astonishing > write speeds from compressible data, and that on random data, it is > significantly worse, so comparing headline write speeds is not that > useful. Yes, but don't read too much into that. It doesn't mean the drive falls over and acts like a 4200RPM PIO ATA drive, it just means the "easy" benchmarking shows compressible numbers, which are higher than what it's likely to get on real data. So with simplistic checking, they often beat other SSD's quite hard on benchmark numbers, but don't really pull off the same crushing in real life. Nowadays a lot of the benchmarks you'll find test them with incompressible data too, and they generally wind up in the same territory as other major SSD's with that. > I've also read that MLC flash is significantly slower writing to the > 2nd bit, and so the drive firmware will constantly be moving bits in > the background from 1st bit to 2nd bit, That... isn't even meaningful. So somebody wrote some stuff on an acid trip ;p > Is it still recommended to have a mirror for log device, now that > pools can survive losing a log device unexpectedly? Remember how the log works; it only ever gets read in the case of an unclean shutdown, so your only chance to lose data is that stuff gets committed into the log, THEN you get a crash/powerpull/etc, THEN the drive dies/throws errors when you come back up. If it dies any other time, no problem, it just disables. The previous state was that once you added a log device, it was inextricably part of the pool, could never be removed, and it dying meant you lost the pool. Now [theoretically] that shouldn't be the case; the worst that can happen is you lost those couple-seconds of data. > The drives I am thinking of getting are either Intel 330, Intel 520, > Crucial M4 RealSSD or Samsung 830, all in their 120/128GB variants. I've got a m4 in the system I'm typing on, and an 830 in the system the typing is happening on. And there's another system (currently in testing, not yet in production) with a 520 in it. I like the m4 and 830; we mostly wound up with the 520 because happenstance made it the cheaper choice in that case. -- Matthew Fuller (MF4839) | fullermd@over-yonder.net Systems/Network Administrator | http://www.over-yonder.net/~fullermd/ On the Internet, nobody can hear you scream.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20121109015711.GB66994>