how and why to benchmark

[Jan E. Schotsman]

-------------------- Begin Original Message --------------------

Message text written by Peter van der Linden

"The disk tutorial I mentioned is at
http://www.acmqueue.org/modules.php?name=Content&pa=showpage&pid=46"


-------------------- End Original Message --------------------

Very interesting link and very readable, Peter.
However, I don't see how anything in this tutorial pleads against disk
defragmentation. Surely the way a disk controller lays out data on disk is
intended to optimize I/O requests? The only model for that is temporal and
spatial locality of data as translated into disk adresses. Anything else
would require a large amount of AI in the disk controller so it may somehow
adapt to particular patterns of I/O requests, which is utterly out of the
question.
As for disk flaws, you aren't suggesting that modern disks keep developing
flaws at the same rate of the infamous Syquest removables, do you? A disk
reformat should solve that problem for some time.

You can't compare the Mac OS virtual memory system to disk layout. The
reasons behind it are very different and speed isn't foremost among them.
Efficient use of all available RAM is one reason, making out of memory
conditions rarer is another.

I don't see why the case for disk fragmentation should be different on X
than on 9 either. 

The main point of the tutorial is that disk layout have moved from system
programmers to the disk controller. Why do you believe that, say, Speed
Disk can do things on 9 that it cannot do on X, assuming it gets all file
access permission it needs?

Jan.

PS. I realize some people were relieved this technical thread had almost
come to an end ;-)