>> Try again with a biger block size.  Say 32k, 128k or 1024k.  I
>> noticed that really small blok sizes like 512 bytes can be horribly
>> slow.

> That certainly helps, with 16K records I get 700+ K/second.
>                              32K records I get 728+ K/second.

> This is also a very interesting metric because it means that at
> "nominal" block sizes, the I/O time is dominated by the MSCP setup
> time.

Are you sure of that?  I've seen performance problems like this with
reason to think they're actually due to each transfer just barely
missing the sector it wants and hence having to wait for a full disk
rotation.  (Of course, depending on how smart the disk in question is,
this may not be plausible.)

					der Mouse

			       mouse@rodents.montreal.qc.ca
		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B