Re: Paging file

[vhamberg@xxxxxxxxx]

James

I'm catching up - remember that we're talking about many
arms, so objects of sufficient size are spread across
multiple disks. This gives good performance because of
the inherent parallelism that gives.

"Sufficient size" means, probably, over 1 meg. The
largest piece of an object that will reside on one arm
is 1 meg, the maximum size of an extent. Smaller objects
may use smaller extents, I believe.

Now in DOS and Windows, you get fragmentation, because
objects are all over the disk as they get changed, so
performance goes down. NTFS is supposed to have less
trouble with this, IIRC. And the EXT2FS of Linux (is
that still used?) is supposed to be pretty good here,
too - it cleans up after itself.

Regards

Vern
> On Tue, 29 Oct 2002, Hans Boldt wrote:
> In addition to this it seems to me that there is another performance
> problem with single level store.  I don't know this for sure but it seems
> right.  If an object in main memory needs to be pushed back to disk
> because memory is low, it is written back to its original location on
> disk.  Then when it is retrieved back into main memory it is read again
> from that spot on the disk.  This is ineffecient use of disk, as these
> reads/writes will be all over the place instead of in a closely clustered
> area of disk.
>
> James Rich