Marcelo Tosatti <[EMAIL PROTECTED]> writes:
> Additional caching of block locations makes read/write (once we have it)
> clustering more efficient because it avoids additional accesses to disk
> (which are going to cause seek time in the worst case).

Are you suggesting changing the block allocation behaviour basd on
which bitmaps are in memory?  That would be an interesting thing to
try, but it is hard to say how it would affect the overall behaviour
of ext2.  (My immediate reaction is that it would tend to increase
file fragmentation, since the disk available for allocation -- that
is, those block groups that have in-memory bitmaps -- would always be
nearly full, at least once the machine reached a steady state).

> Another thing you can try is to make the buffer cache shrinking (called
> when memory shortage happens) less aggressive. (search for
> kmem_cache_reap() on mm/)

Yes, one possibility is to give block bitmap buffers a priority based
on the fullness of earlier block groups.  But that would involve more
delving into the 2.4 mm code than I've done up till now.

(Also, if I was doing something that sophisticated I'd probably want
to track Al Viro's work on moving ext2 to use the page cache rather
than the buffer cache throughout.)

> > One possibility is that ext2 is the simply the wrong filesystem for my
> > application, and any other applications that needs guaranteed
> > sustained write bandwidth.  The obvious alternative is XFS.  However,
> > early releases of Linux XFS will not feature the guaranteed rate I/O
> > features, so that's not likely to be an option for quite a while yet.
> 
> Even without guaranteed rate IO XFS extents will probably make the
> throughtput higher.

Apart from the block group bitmap read issue, ext2 seems to be getting
near to the maximum write throughput I would expect for the drives I'm
experimenting with, so I doubt XFS has much room for improvement.

The slide at
<URL:http://oss.sgi.com/projects/xfs/papers/usenix2000/img13.htm>
seems to confirm this.  Of course, those results are at least 6 months
old, and XFS for Linux is still under development, so I'm not
discounting it yet.


David Wragg

-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to