Subject: Re: SCSI Target Mode
To: Brian Hechinger <>
From: Roger Ivie <>
List: port-vax
Date: 08/19/2004 17:19:33
On Thu, 19 Aug 2004, Brian Hechinger wrote:
> does anyone remember the discussions we had a ways back about this?  mostly
> our purposes were to get IP working on SCSI to use the SCSI busses of the
> VAXstation class machines as network interfaces.
> is this possible?  this would solve a couple other situations for me as well,
> like using VMS, where SATA is as of yet unsupported.

It's certain *possible*. I once did this on VMS between a /90 and a /60.
I had DECnet and TCP/IP running over the interface.  However, I did it
using the dual-SCSI TURBOchannel interface (PMAZ-C) instead of the
motherboard controller. I needed the motherboard controller for the
system disk in both systems.

Bear in mind that it will NOT out-perform Ethernet. The difficulty with
SCSI is that there are several places during the transmission of a
packet that an interrupt needs to be taken; in my implementation, a
transmitter had to take (IIRC) three interrupts and a receiver had to
take two. An Ethernet interface can slurp a bunch of packets into memory
with a single interrupt. Interrupts are expensive; I discovered that the
/60 responds to interrupts MUCH faster than a /90 does.

Although the data transfer itself was much faster than going over the
Ethernet, the interrupt overhead brought the performance down to Ethernet

The TURBOchannel adapter is not fast, so it's possible that the
motherboard SCSI would be faster. However, since the critical timing is
in dealing with the interrupts it is not likely to make a difference.

I needed the dual-channel interface to deal with the possibility
that I might be addressed to receive a packet while I was preparing to
send one; I just couldn't figure out how to handle that case with a
single chip.
Roger Ivie
Version: 3.12
GCS/P d- s:+++ a+ C++ UB--(++++) !P L- !E W++ N++ o-- K w O- M+ V+++ PS+
PE++ Y+ PGP t+ 5+ X-- R tv++ b++ DI+++ D+ G e++ h--- r+++ z+++