Matthias Bertschy wrote: > Hello, > > I am in the process of creating a server codenamed "terabyte server" > which will hold some of our backups. > The setup is fairly easy: a RAID card and four 300G ATA drives which > will make approx 1.2Tbyte.
actually, by the time you get done with marketing-speak, you are closer to 1T (OpenBSD's capacity isn't in marketing-speak). Of completely non-redundant, 4x failure point, new technology disk space. Be scared. (math: the 300G drives I've seen are about 280G. 280Gx4 = 1120G 1120G / 1024G/TB = 1.09TB. Maybe your "300G" drives are bigger than the ones I've seen...don't count on it) > Of course I chose OpenBSD 3.8 for it. During the installation > process, I realised my biggest problem: OpenBSD doesn't want to make > file systems as big as 1.2Tbyte. > > In fact, I did my homework and read the FAQ 14.7 which states > "OpenBSD supports an individual file system of up to 231-1, or > 2,147,483,647 sectors, and as each sector is 512 bytes, that's a tiny > amount less than 1T." > > Ok, well, I am not a system programmer, but would it be difficult to > make an OpenBSD support for 2Tbyte file systems VERY difficult. Think about the consequences of even the smallest error. Think about how you test EVERY IMAGINABLE edge case. Think about how long it takes to fix things when your test 2TB disk array gets hosed. Again. [totally OT: ever think about how much frustration Marco and crew had to go through to get all that cool bioctl stuff working? "DANG, that didn't work, time to reboot the server...AGAIN!" Rebooting those servers he was working with is a most unpleasant task...] Yes, it has to be done, but it will be done when someone does it. You don't expect I (or anyone else) can give you a date, do you? I can pretty well tell you it won't be in 3.9, as such a beast would probably have to be committed the day after unlock (long before you get your CDs) in order to be sufficiently tested for the next release. I can assure you, you will know when it happens. > like on Linux? oh, thrash from File-system-of-the-week to file-system-of-the-week? that might not be so difficult. (ok, cheap shot, I know, couldn't resist). > (for > the moment "terabyte server" runs Linux, but I would prefer OpenBSD) A story: Long ago, when the first 1G drives came out, they were a small hair bigger than 1G in size. The way MSDOS handled them, because of that hair, they went to the largest block size (32k). Throw away a tiny amount of space (I think it was something like 10M or less), MSDOS would slip to a 16k block size, and the user would end up with effectively considerably more usable space. However, people would see me partitioning down their disk and would say, "I PAID FOR IT, I WANT IT ALL!", or insist that I give them a 10M D: partition. (Worst case I saw was a client with a full 200M HD, lots of tiny files. She got a new computer, with a "huge" 1G disk, copied her old disk over and she ended up with...200M of free space, all due to this block size issue). The point: Sometimes people obsess over the silliest things, and hurt themselves in the process. For 1.1T of disk space, either break it down to two 550M chunks, or just throw away the 100G. (or use it as a separate "test" partition) You won't miss it...at least, until one of those drives fails, and you wish you threw away an entire 280G and went with redundancy. "But! I want it all in one file system!" Ok, so what do you do when the 1.1TB becomes insufficient? You are going to buy four new, bigger drives, and another controller and copy everything to the new drive and throw away the old ones? Probably not: you will probably just add some more storage and make it two file systems. Just do that now. :) Seriously, if you can't manage multiple blocks of data now, odds are you are going to be in big trouble when that "big" system starts looking small. IF you really think you won't outgrow that "big" 1.1TB system and don't have to worry about add-on storage, you can certainly throw away 100G and run your prefered OS. I'm really curious how you are going to back that thing up, too. 4x failure points and relatively cutting edge disks...it better be GOOD. Nick.
