Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!snorkelwacker.mit.edu!bloom-picayune.mit.edu!athena.mit.edu!mmshah From: mmshah@athena.mit.edu (Milan M Shah) Newsgroups: comp.sys.ibm.pc.hardware Subject: Re: Drive Arrays Message-ID: <1991Apr9.210740.16385@athena.mit.edu> Date: 9 Apr 91 21:07:40 GMT References: <46563@ut-emx.uucp> Sender: news@athena.mit.edu (News system) Distribution: na Organization: Massachusetts Institute of Technology Lines: 58 >solomon@chaos.utexas.edu (Thomas Solomon) writes: > > >> I recently received some product information from Dell computers. >>They claim to have a controller that they refer to as the "Dell >>Drive Array" that can handle up to 10 disk drives at the same time? >>I get the impression from their information that this controller can >>actually write to these different disks _at the same time_, allowing >>blazing fast disk access. > >Both can "data stripe" to multiple drives simultaneously. Your data is >broken up and stored on several drives. This means you're not limited >to the transfer rate of a single drive. You're limited to the transfer >rate of the EISA bus which is 32-bits wide. > >Both provide data guarding. On the Dell drive array, your data is >striped on four disks with a fifth for parity. All this should be taken with a grain of salt. First of all, it is not correct to expect that with n drives, your I/O system becomes n times faster than with 1 drive. It seems that the Dell drive uses the fifth drive for parity. This means that every write must write to the parity drive, so you are still limited by the data transfer rate of a single drive. So, even if sector 1 and sector 2 reside on different drives, I must access them sequentially cause I have to access the parity drive for each. Actually, a better idea is to spread the parity across all drives. ie, parity for sector 1 is stored on disk 1, for 2 on 2 for 6 back to 1 etc. Even then you are limited. Lastly, if you break up the time involved in a typical I/O request, one finds that about 8ms go for seek, about that much on rotating the disk so that the sector is under the head, and only about 2ms in actually transferring the data. With a single drive, on average you will have to wait for half a rotation to find your sector; with n drives, you tend to have to wait for one entire rotation (unless the drives are synchronized). You might actually loose here! This stuff is actually enough for a Master's thesis (mine, I hope!!!) and therefore can't be covered in any detail here. For an excellent background, read David Patterson's and Katz's papers on RAID - Redundant arrays of inexpensive disks. I believe that to a very crude first approximation, the merits of drive arrays lie in the fact that a 2 gigabyte store made up of 10 200 Meg drives will be cheaper, more reliable and *maybe* faster than a single 2 Gigabyte drive. It is incorrect to assume that a 2 gigabyte store made up of 10 200 Meg drives will be 10 times faster than a single 200 Meg drive. But of course, I am ready to bet every last dollar I have that the marketing types at Dell and Compaq and IBM etc., will convince one and all that in fact their arrays are 10 times faster, and will sell a ton load of them, leaving a trail of puzzled and appropriately impoverished customers behind. Hey, if 4 times oversampling in CD players is such a good thing, drive arrays of 10 gotta be good too, eh? ;-) Milan .