Bus performance (SCSI & S-Bus)

[Steven Gutfreund]

I have some questions on the S-Bus and SCSI bus performance:

SCSI:

Why do the spec sheets claim that an ASYNC SCSI runs at 1.5Mbyte/sec and a
SYNC SCSI at 4.0Mbytes/sec (I asssume peak). I understand how a
synchronous protocol could up total aggregate throughput when several
devices are contending for the SCSI (given sufficient cache).  But it is
mystifying how one could claim that the ASYNC protocol is so poor that one
SCSI disk could run more than twice as fast with the synchronous protocol.

S-Bus:

Bechtolsheim claims in his article at 100Mbyte performace for the S-Bus (I
assume this is gotton via by multiplying the 25Mhz bus clock speed by a 32
bit transfer unit). [let's call this peak]. But I have seen articles here
that the SparcStationI is only running at 29Mbytes.  However, the
SparcStationI does have a 25Mhz clock on the bus (according to spec sheets
I have seen). Furthermore Sun tells me that its peak cache fill rate is
25Mbytes/sec.

I am quite confused. But basically what I want to know is if there is any
limitations on the SparcStationI bus master (processor board). That a swap
with the SparcStationII processor board will mean greater perforamce.

A Gordon Bell quote (he spoke here yesterday):

When we (Stardent) quote you a Mflop, MIP, Linpack peak figure what it
really means is that we GUARANTEE you will never ever exceed that number,
what you really get depends on how well your application keeps ALL the
pipes full.

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Yechezkal Shimon Gutfreund		 		  sgutfreund at gte.com
GTE Laboratories, Waltham MA			    harvard!bunny!sgutfreund
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=