[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: announce: readahead-1.4



On Wed, 2007-03-07 at 12:55 -0500, Tony Nelson wrote:
> Microsoft already addressed that issue in WinXP, which has fancy hooks to
> the disk defragmenter so that there isn't much seeking when booting WinXP
> (it does take several days until the defrag happens).  So I agree with
> Callum, and also dunno WTR MS is thinking.
> 
> Probably readahead adopting what MS did in WinXP would be most effective
> but would also violate the most MS patents, as well as requiring hooking
> into the filesystem rather more than is wanted.  Something like UnionFS
> might work without patent issues, if it could use a file instead of a hard
> partition.

I don't see how they could patent defragging a disk. Lets not get crazy
here. ext3 does a decent job of not fragmenting files unnecessarily, can
we really gain much more than the current readahead solution?

Instead of reaching for more and more convoluted hacks, lets do it
right. Seems to me the only real gain to be found is in having
performance critical files clustered together in the start of the
partition, where the disk is typically faster.

Food for thought, check out Apple's HotFile implementation:

http://developer.apple.com/technotes/tn/tn1150.html#HotFile

Doing something like this on ext3 should be possible.

We could designate the first ext3 block group (actually should make the
number configurable) as the "hot" area. Patch the kernel so it avoids
doing any allocations in the "hot" group(s) except as a last resort.
Then write a tool to move files into the "hot" area somehow, possibly
with help from the kernel.

Attachment: signature.asc
Description: This is a digitally signed message part


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]