jmcgrath at qntm.com
Fri Oct 13 13:32:22 PDT 1995
Reply to: RE>>LVDS question
To elaborate on Skips's point, SCSI only requires about 15 us of
bus overhead for a single IO (command, disconnect, reconnect, data,
status). Anything more means someone's software or hardware is
slowing things up - and those are NOT under the control of the standard.
For 2K commands, 15 us is big - but short, random commands cannot
saturate a SCSI bus (the drive IO/s rate is far too low). FOr large
commands you can get bus saturation, but there the bus overhead
due to protocol is a very small percentage - you are better off making
the data transfer time shorter.
Date: 10/13/95 1:01 PM
To: Jim McGrath
From: Skip Jones
This topic has been discussed several times. I share the view with those that
believe bad (excessive) overhead in SCSI has more to do with poor
implementation than what is inherent in the standard. If implemented well
'good' silicon and well written firmware/software, SCSI overhead is already
fairly low at under 200 microseconds for the entire I/O (host AND target
included), and in some cases much less.
Therefore, I view this effort as a lot work for little gain. Poor
implementations will persist regardless of what we do in the standard.
From: Lee Cleveland on Thu, Oct 12, 1995 8:04 AM
Subject: LVDS question
To: SCSI at symbios.com
First Thanks for all the good dialog on LVDS
Is it being concidered to change some of the SCSI overhead/arbitration
signaling when going to LVDS?
If so is there a list of what is being looked at?
IBM Rochester MN
IOP Design & Hardware Development
More information about the T10