Messages in this thread | | | Date | Thu, 15 Jul 1999 23:03:36 -0700 | From | merblich <> | Subject | Re: Large disk partition over 300GB |
| |
Gotta disagree somewhat ... yes time vs space :) The logic was right and not quite 10 years...
I remember back in '95 on Solaris 2.5.1 oops, yes, Solaris that it supported at least in some fragment reallocation function the support for time vs space based on a minfree value...This is/was ufs code.
Ask your Sun buddies to check the Solaris 8 code, would not doubt that it is still there..
I haven't looked at ext2 code.
Mitch PS: Got one for you, what does ABR mean? Hint it is a QoS ATM term... =======
Stephen C. Tweedie wrote: > > Hi, > > On Wed, 14 Jul 1999 22:16:18 -0700, merblich <merblich@gateway.net> > said: > > > Normally, a file system will allocate disk blocks based on a time > > vs space algorithm. By having blocks free in a number of locations, > > the best location to allocate the disk block due to the growing > > of fragments or just a new block can be realized. A couple of factors > > in determining this is interleave, rotation speed, time to seek, etc. > > Not any more. That used to be the space, nobody makes drives like that > any more. For the last ten years there have been large numbers of > pipeline stages added into the disk IO path, from track buffers to > on-disk caches to asynchronous scsi tagged commands to operating system > readahead buffers. The _only_ things you can effectively optimise for > when allocating data are (a) sequential access, and (b) approximate > proximity of related data. Effective explicit data placement just > cannot be done by the OS any more, and ext2 (being less than ten years > old) has never tried to do so. > > > Once the file system's free space decreases to a specified point, the > > allocation then just allocates blocks due to their availability. My > > guess, is that 5% is the point where this tradeoff is made. > > Not at all. Ext2 always uses the same algorithm. It just becomes less > effective if there is less free space. The exact rate at which things > fall off obviously depends on the access patterns, which is one reason > why the reserved space region is configurable. > > --Stephen
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |