Amanda-Users

Re: Garbage in tape restore

2002-08-03 02:05:22
Subject: Re: Garbage in tape restore
From: Gene Heskett <gene_heskett AT iolinc DOT net>
To: Terry Koyama <terrykoyama AT yahoo DOT com>, amanda-users AT amanda DOT org
Date: Sat, 3 Aug 2002 04:51:03 -0400
On Friday 02 August 2002 23:59, Terry Koyama wrote:
>All,
>
>Finally solved the problem.  It was the SCSI Cable.
>Before trying a different SCSI cable, I
>transferred the SCSI controller and tape drive to
>a Windows 2000 machine.  I updated the firmware
>and ran all of its diagnostic tools.  Everything
>ran fine.  I even made a few backups with no
>problem.  Remember, this was with the original cable.
>
>I then transferred the SCSI controller and tape drive
>back to the original Linux machine (with the original
>cable.  Still the same problem.  So I asked a
>friend to borrow his SCSI cable and poof...it worked!

The original's warranty, probably written in Rhode Island Red or 
Plymouth Rock chicken blood, ran out.  Those things need a warranty 
extension via the usual chicken sacrifice from time to time.  
Really stubborn cases will require a goat.  :-)

On a more serious note, scsi cables are by nature a 'transmission 
line', with their perfomance and history firmly buried in classic 
transmission line theory.  And as such they must be terminated in 
their characteristic impedance at the physical ends, and *only* at 
the physical ends of the cable.  This impedance is typically for 
flat cables, about 120 ohms and is implicite in the scsi spec 
primarily because that was the characteristic of the flat cable 
available at the time (and still is).

The usual resistive terminator package gives about 132 ohms as its a 
quasi-parallel combination of a 330 ohm going to ground, and a 220 
ohm going to the termination power supply, usually a diode drop 
below the 5 volt line, usable but not perfect, and totally blown to 
hell by someones plugging in the term packs swapped end for end, 
thereby reducing the standing voltage for a logic 1 from 3.0 volts 
to only 2 volts which totally destroys the noise margin of a TTL 
circuit.

A lashup that uses 'active' terms will come quite a bit closer since 
the buildout resistor there is usually 120 ohms on the money.

The length spec allows many meters of cable in a properly setup 
system, but if the last device is a foot from the end of the cable 
just because the connector location was handy, that last foot of 
unterminated cable is going to ring like the liberty bell did, once 
for every signal transition.  Folks who don't understand that can 
send for the chickens and the voodoo spells, but they'll be wasting 
their money.

I've also had relatively poor luck with the so-called high quality, 
round (because they are all rolled up) cables, and have learned to 
replace them on sight if there is any hint of scsi bus problems.

One of the toolkits a techie needs is a roll of scsi cable in the 
correct width and a bag of connectors to match, which in 10 minutes 
will make you a better cable than 99% of what you can buy at 
Staples et all, and for 20% of the cost of the blister-packed 
versions.

Then, when all good design practices have been put in place, is the 
time to sacrifice that goat.  :-)

[...]

-- 
Cheers, Gene
AMD K6-III@500mhz 320M
Athlon1600XP@1400mhz  512M
99.09% setiathome rank, not too shabby for a WV hillbilly

<Prev in Thread] Current Thread [Next in Thread>