>   I also suspect that tweaking the file system parameters probably
> isn't going to make a whole lot of difference in the overall performance.
> A lot of the I/O is likely to be small transfers (512, 1K, 4K, or 8K),
> and if there is more than one process doing I/O, the I/O will be scattered
> over the disk.

True.

Is there any hope of getting dynamic disk buffering to NetBSD in the near
future?  I mean the system that uses all free RAM for disk buffering,
instead of some relatively small fixed-size buffer.  A fixed-size buffer
would be OK if we all had some 16 MB of RAM for that purpose alone.. but
on A3000, that's all we've got :-I

Another vast improvement (I believe) would be a disconnecting SCSI driver
for the A3000 (one that doesn't busy wait in kernel during all disk
operations).

Is anyone working on these?

  Jukka Marin


P.S. Any A2232 users out there?  Can you make the RTS handshaking work?
     I may rewrite the 65C02 code when I have the time..