Date: Sun, 05 Oct 2014 23:37:18 +0200 From: InterNetX - Juergen Gotteswinter <jg@internetx.com> To: FreeBSD Filesystems <freebsd-fs@freebsd.org> Subject: Re: Mirrored SSDs for ZIL/SLOG - safety, flushing, capacitors Message-ID: <5431BA0E.7090707@internetx.com> In-Reply-To: <864mvigtxd.wl%poyopoyo@puripuri.plala.or.jp> References: <CAP1HOmRYy8cX9k7rs4UEzx41Og4phX5U_2OpiUCKBDSRdMf_bQ@mail.gmail.com> <alpine.BSF.2.11.1410030547070.32407@wonkity.com> <864mvigtxd.wl%poyopoyo@puripuri.plala.or.jp>
next in thread | previous in thread | raw e-mail | index | archive | help
Am 05.10.2014 um 22:09 schrieb poyopoyo@puripuri.plala.or.jp: > At Fri, 3 Oct 2014 05:59:48 -0600 (MDT), > Warren Block wrote: >> I think the Crucial M550 SSDs have additional backup capacitors also. >> However, I'm still skeptical about the value of a small SSD power backup >> versus a true UPS for the entire system. > > I have read an interesting article on this topic: > > http://www.anandtech.com/show/8528 > Micron M600 (128GB, 256GB & 1TB) SSD Review > (The Truth About Micron's Power-Loss Protection) > > The latter part of this article reports Micron's position of this > specific function. In my understanding, Micron's consumer-class SSDs > such as M550 have very small backup capacitors which would not be > worth flushing dirty DRAM cache, but completing ongoing NAND > programming. This means while programming at power loss would not > corrupt NAND, "successfully written" host write would be lost due to > cache loss. It does not provide enough peace of mind required for > logging device. > > It looks like metadata journaling or softupdates for filesystem, not a > full journaling implementation. > Nexenta for example recommends Stec SSD, SLC Type. Since the Zil is going to be hammered with writes all the time (mirroring adds another brick to this), its probably a good idea avoid any consumer grade mlc type ssd. look for nexentas / oracles / suns hardware compatibilty lists, to get a clue which brand/type fits good for this job. we are running a nexenta ha cluster since ~3 years, equipped with dell branded pliant lb150s slc ssd (plaint is sandisk nowadays), absolutly flawless. all 4 are still the initially installed flash drives. i dont think that any 150€ ssd will take this for mid/long term, at least performance degradation should start very quickly under zil/zfs conditions. also keep in mind that the zil dont need to be such big, most of the time 16-32gb are more than enaugh. you wont take any profit from huge zil drives. but, like kevin already mentioned before, you really should think about your usecase and if you need dedicated zils. l2arc can be done with every crappy ssd out there, since it wont affect your data consistency if one dies. which protocoll are you going to use to export your shares? ifs its going to be a nfs filer, you should investigate further into this before since you will most likely see pretty bad write performance on the shares and recommendings to disable zil or patch the nfs server to become async. this makes dedicated zils nonsense... cheers, Juergen
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5431BA0E.7090707>