On Thu, Jun 07, 2001 at 09:46:25PM +1000, Steve Kowalik wrote: > On Wed, Jun 06, 2001 at 04:13:09PM -0500, Andrew D Dixon uttered: > > What I'm trying to figure out specifically is if I have a file, or a > > directory, at the beginning of my disk and it start to get bigger and > > bigger will it push the rest of the file system towards the end of the > > disk or will it just write all of the new bits at the end of the disk? > > > Writing to a drive does not work like this. > All a directory is a list of inodes. (In the case of ext2, anyway), so if > you add files to the directory, the new inodes (which can, and usually are, > all over the disk) will be appended to the list.
Erm, no, that's not how it works at all. A directory is not a list of inodes. A directory is a file containing directory entries. A directory entry is not the same thing as an inode (for example, inodes don't have names, the way files do. Multiple directory entries can have different names but point to the same inode). Inodes are also not "all over the disk". (except in NetApp's WAFL filesystem). There is a specific section of the disk allocated to inodes, and it is static and not a part of the filesystem at all. Once this portion is created it can't (easily) be changed. You are stuck with the number of inodes you've created. What the person is asking about is not "garbage collection" but fragmentation. They are asking, basically, whether or not a file that grows (syslog, for example) will get fragmented when it tries to grow but runs up against a block that has already be allocated to another file. The answer to their question is yes, the file will get fragmented. However, the way Unix filesystems work (or any non-FAT FS) is that they take steps to minimize fragmentation. Often this is done by over allocating space for a file. The average file on a Unix system is something like 4 KB, so the FS allocates more than that initially. When you boot your linux system, fsck will let you know how fragmented the filesystem is (it will say something like "8.4% non-contiguous"). This number will typically be small, less than 10. Only if the disk becomes very full do you risk a lot of fragmentation, and that's because there's no longer enough space for the FS to leave gaps between newly created files, so they may not be able to grow much at all before fragmenting. As it turns out, it's often better to have large files fragmented anyway. I don't remember why this is, but I suspect it has something to do with the fact that they'll wind up on different drive heads and will have the effect of actually speeding up seek times. I could be completely wrong about that, though. noah -- _______________________________________________________ | Web: http://web.morgul.net/~frodo/ | PGP Public Key: http://web.morgul.net/~frodo/mail.html
pgpuRvnTRwmgc.pgp
Description: PGP signature
