On Thursday 08 March 2007, Alan Pearson wrote:
>On 8 Mar 2007, at 19:12, Michael Loftis wrote:
>> SCSI is backwards compatible, an U320 LVD device will work on a U80
>> LVD controller, just at U80 LVD speeds. I'd suggest just getting a
>> U160 or U320 controller, and make sure to get *good* SCSI cables.
>> That's partly what bit me in the butt this last time. I ended up
>> with an adaptec 2944, which has been pretty solid.
>
>I'll second the cables recommendation, don't skimp on them, you'll
>pay big time with daft problems and errors !
>
>On the card, I use adaptec here on Linux (U320 PCI-X) and it's never
>given an issue. It drives a LTO3 drive at about 65Mb a sec... nice !
>I understand though that Adaptec can be very funny about their drives
>as Mr T. De Radlt from OpenBSD has had many arguments with them ...
>
>
>---
>AlanP
Proper termination is the scsi equ of an achilles heel, and lots of
virgins have been taken to the alter over that since scsi was invented.
Few folks realize that any cabling handling data transitions that are
usually at 10ns or less rise and fall times is not 'just a cable' but is
in fact a transmission line, subject to all the rules that apply to the
successfull use of transmission lines to move power or data any great
distance when compared to the distance the signal can travel in the time
of that transition. The common ribbon cable, built in .050" centers,
assuming that every other wire in it is a ground, has an operating
impedance one can calculate from the textbook formulas that is in the 124
ohm region. This is why the original scsi stuff was set up to use
powered terminations, using a 220 ohm resistor from each signal line to
the 5 volt buss and another 330 ohm resistor from that signal line to
ground. That combination has, by parallel ohms law, an effective
resistance of about 132 ohms and is close enough to the wire impedance to
serve as a termination while simultainiously setting the bias on that
wire (remember, this is an open collector buss, where any host or client
can pull the line down, but none can pull it up, only theterms can do
that) to around 3 volts, which is comfortably above the TTL logics 2.4
volts that is guaranteed to be seen as a logic 1. Thats about a 600
millivolt cushion which can absorb some ringing from poor terminations
and still work 100%.
Now, in order to keep an external scsi box with its own termination supply
from powering the host computer via the term power line when it was
turned off, a diode was inserted to prevent this cross coupling. It was
originally thought that a schotkey would be used as its voltage drop is
usually less than 200 millivolts, so you still get around 2.85 volts on
the bus lines when quiet. But you've now lost 150 millivolts of this
noise cushion, so things have to be a little closer than the 20%
resistors used in the term packs can guarantee 100% of the time, so the
occasional virgin gets sacrificed to make it work.
Then some bean counter at adaptec et all sees the expense of that $1.27
schotkey diode and crosses it off the list, substituting an si diode for
$0.09 as it goes by him on the way to production, not knowing or caring
that his 'minor' change just screwed up an engineering design that was
already exporing the edges of the envelope. Then that 200 mv loss
becomes as much as 750mv, minimum 620mv, and the logic one resting
voltage on the buss is now 2.50 to 2.55 volts. You only have a 100 mv to
150 mv noise margin left, and folks, it ain't gonna fly cause you can't
find that many virgins no matter where you advertise. I have personally
ripped out that si diode and put in a schotkey several times, and I've
converted a very flaky scsi buss into one that was bulletproof just as
many times.
The only possible salvation is to pitch the term packs from your design,
and use an active terminator chip instead, one that CAN set the buss at
2.9 or more volts while maintaining a src termination of around 125 ohms
at the same time while running on 4.25 volts as its power source. It
has the side advantage that it only uses any great amount of power when
a data line is pulled to a logic zero compared to the termpacks sucking
down most of a watt on both ends of the cable and going toward a couple
of watts when the bus was busy. It took a while for that technology to
become competitively priced but I think its a lot more universally used
today than 15 years ago.
Getting this problem out of sight and out of the consumers mind was first
done with the IDE bus where the terms are hidden but there nonetheless,
and now to SATA where they've gotten even more precisely controlled is a
good thing IMO as long as nobody gets cute with the cabling. If you do,
then you're back to advertising for virgins.
Yeah, I'm a C.E.T. who's been in broadcast engineering for 45 years, I
_know_ about vswr related things first hand. We fight and do sometimes
heroic things to get that miss-match down to fractions of an ohm.
--
Cheers, Gene
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
"I only touch base with reality on an as-needed basis!"
-- Royal Floyd Mengot (Klaus)
|