LVDS question

Lohmeyer, John JLOHMEYE at cosmpdaero.ftcollinsco.ncr.com
Thu Oct 12 09:43:00 PDT 1995


Lee wrote:
 ----------
>Question:
>  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?

I brought up the subject at a recent working group meeting and the idea 
received a rather cool reception.  The major reason was concerns on 
interoperability.  To shift into a lower-overhead mode of operation will 
require some kind of negotiation.  People were concerned that however we 
initiate such a negotiation could "break" existing devices.  People still 
remember a lot of hosts that hung when a target attempted to send the 
Synchronous Data Transfer Request message.

A second objection was that SCSI overheads are already quite low and any 
improvements may not impact system performance significantly.  The effect of 
overhead on large I/Os (such as done on a video server) is negligible.  On 
small I/Os (such as transaction processing) the SCSI bus is nowhere near 
saturation, so reducing overhead would not appreciably increase I/Os/second.

Theoretical minimum overheads for a SCSI I/O process are around 18 usec 
(assuming an initial connection to transfer the command and any outbound 
data plus one reconnection to transfer any inbound data and the status). 
 Most recent protocol chips have overheads in the 50 - 100 usec range. 
 System drivers can add considerably more overhead.  So attacking the 
protocol overhead may not be the best place to focus to achieve significant 
gains.

All that said, if we were to pursue this idea I think the biggest 
improvement could be obtained by negotiating a smaller Bus Settle Delay 
(currently 400 ns).  The BSD needs to be set to the round-trip propagation 
time on the bus.  The current BSD was set to allow ~100 feet cables based on 
a rather conservative assumption of 2 ns/ft propagation speed (2 x 100 x 2 = 
400).

Since a BSD occurs between each information transfer phase plus in several 
other places in the protocol, reducing it would likely yield the biggest 
improvement (albeit at the expense of reducing the maximum supported cable 
length).

John
 --
John Lohmeyer             E-Mail:  john.lohmeyer at symbios.com
Symbios Logic Inc.         Voice:  719-573-3362
1635 Aeroplaza Dr.           Fax:  719-573-3037
Colo Spgs, CO 80916     SCSI BBS:  719-574-0424 300--14400 baud




More information about the T10 mailing list