>
>For reference sake, I'm able to get 18.5MB/sec on a SparcStation 10
>and the same performance on a 486 w/PCI running Solaris X86..
>

When you make these kind of comparisons, you should factor
CPU utilization as well as native hardware. Even then,
the relevant host adapters should be stated.

For example, if you compare a DX2 486 (at 66MHZ) running
a script engine NCR on a 133MHZ PCI bus connecting to a
fast wide SCSI bus at 18.5MB/sec, are you comparing it to
the SS10 (with 50MHZ CPU modules?) on a 100MHZ SBus,
using the ESP chip (which requires 8-10 interrupts/command
with CPU intervention each time), or are you comparing it to,
say, the ISP (QLogic microengine)?

Further, 90% utilization of a fast wide SCSI bus (I assume that's
what you're running), of sequential striped disk activity with a
fairly wide stripe is more or less a no-brainer. It would be
more interesting to see what your TPS numbers are, and the
relative CPU costs to get them.