Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.2 9/5/84; site baylor.UUCP Path: utzoo!watmath!clyde!burl!ulysses!gamma!epsilon!zeta!sabre!petrus!bellcore!decvax!genrad!panda!talcott!harvard!seismo!ut-sally!ut-ngp!shell!neuro1!baylor!peter From: peter@baylor.UUCP (Peter da Silva) Newsgroups: net.database Subject: Re: UNIX + database Message-ID: <573@baylor.UUCP> Date: Thu, 29-Aug-85 12:59:54 EDT Article-I.D.: baylor.573 Posted: Thu Aug 29 12:59:54 1985 Date-Received: Wed, 11-Sep-85 06:59:39 EDT References: <164@3comvax.UUCP> Distribution: net Organization: The Power Elite, Houston, TX Lines: 39 > Some people say that 'extent-based' file systems are inherently > faster than unix's. Depends on the file size. If by an 'extent based' system you mean a linked list of large extents, as opposed to a tree of small extents which is what the UNIX file system actually is (an extent is just a single indirect block, in UNIX terminology), then for small (1-2 extent) files UNIX will be slower. Once you have more than 2 extents you're going to be doing 3 indirects with your extent based system anyway, whereas UNIX will still be accessing the first or second indirect block (I'm not sure of the terminology): Assuming buffering and equal blocksize: N = #ptrs/block UNIX: INODE 10 * ^block 1* ^indirect block ( of ^block) 1* ^double indirect block ( of ^indirect block) 1* ^triple indirect block ( of ^double indirect block) E-SYS: EXTENT -1 * ^block 1* ^EXTENT So the first 10 blocks take 1 indirect in each. The next -11 blocks take 1 indirect in your extent system, 2 in UNIX. The next 9 blocks take 2 indirects in each. Then there are another -11 with 3 indirects. After that UNIX is ahead... 3 indirects for UNIX, 4 or more for extents. That may be screwed up somewhere, but by inspection you should be able to see that after the third extent UNIX is ahead. I wonder if that's why this damn IBM-PC is seeking all over the place? Nah. My files aren't that big. -- Peter (Made in Australia) da Silva UUCP: ...!shell!neuro1!{hyd-ptd,baylor,datafac}!peter MCI: PDASILVA; CIS: 70216,1076