Date: Fri, 28 Nov 2008 14:45:31 -0500 (EST) From: vogelke+software@pobox.com (Karl Vogel) To: freebsd-questions@freebsd.org Subject: Re: 5 TB server Message-ID: <20081128194531.7B17FB7C5@kev.msw.wpafb.af.mil> In-Reply-To: <139b44430811280548x36915301i766bfb15f162c8ca@mail.gmail.com> (valentin.bud@gmail.com)
next in thread | previous in thread | raw e-mail | index | archive | help
>> On Fri, 28 Nov 2008 15:48:45 +0200, >> "Valentin Bud" <valentin.bud@gmail.com> said: V> I have to come up with a solution for a company that has as we speak 4 TB V> of data spread among 3 computers with lots of HDDs. Of course I've V> recommend them to buy a server for that storage capacity and for data V> organization. Good idea. We had a similar problem here; a SuperMicro server tanked, and the company that provided the warranty went bankrupt, so all we had were 12 perfectly good 400-Gb SATA drives and nowhere to put them. The drives were the big-cost item, so I didn't want to just dump them. We finally bought two empty IBM x3400 8-bay enclosures plus some IBM SAS 3.5" hot-swap trays, and it's working like a charm. V> I thought of going on the ZFS way (on FreeBSD of course) with some raidz. In my experience, completely new filesystems or operating systems need at least 5 years in the field to weed out all the weird corner-cases. I might trust ZFS on Sun hardware (*with* vendor support) at this point, but I'd wait awhile before trying it on anything else. This isn't a slam at ZFS or the FreeBSD porters, it's just recognition of the fact that some types of software development are *not* time- compressible, regardless of who's doing the work. V> One of the problems is that the server will stay in their office so it V> has to be quite silent. Not a good idea, especially if this data is their bread-and-butter. You can walk out the door with a system this size on your shoulder, so I'd recommend a locked room with reasonable cooling and *clean* power. You don't need a 10-foot-tall zillion-dollar Liebert A/C, but you absolutely need a UPS that can take care of power spikes; the more moderate the environment, the less likely you are to have a hardware failure. I don't use disk mirroring because 99% of our problems come from humans rather than hardware. If someone zaps the wrong file, a mirror will simply replicate that mistake; we have two matching servers in separate rooms, and we run rsync nightly to back up the production box without deleting any files. I also run hourly backups on the production box to store anything that's been modified in the last 60 minutes, which gives us a nice file history and takes care of most recovery problems. With two servers, I can use basic UFS filesystems and get fine I/O performance with minimum maintenance. -- Karl Vogel I don't speak for the USAF or my company The first product Motorola started to develop was a record player for cars. At that time, the most known player on the market was Victrola, so they called themselves Motorola. --possibly-true item for a lull in conversation
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20081128194531.7B17FB7C5>