tape drive error
john bray
jmblin at comcast.net
Wed Mar 16 20:55:14 UTC 2005
gene -
thanks for that wonderful set of thoughts!
john
On Wed, 2005-03-16 at 13:56 -0500, Gene Heskett wrote:
> On Wednesday 16 March 2005 10:25, Eric Shibata wrote:
> >Hi Gene,
> >Do you mean interrupts?
> >When I look at my /proc/interrupts I know I have other things there
> > with my scsi adapter from BusLogic.
> >
> >--------------------------------------------------------------------
> >---------------------------------------------- CPU0
> > 0: 70937522 XT-PIC timer
> > 1: 200 XT-PIC i8042
> > 2: 0 XT-PIC cascade
> > 8: 1 XT-PIC rtc
> > 11: 368831 XT-PIC BusLogic BT-930, uhci_hcd, eth0
> > 12: 4778 XT-PIC i8042
> > 14: 90206 XT-PIC ide0
> > 15: 212674 XT-PIC ide1
> >NMI: 0
> >ERR: 0
> >-----------------------------------------------------------
> >Let's just say it was a cabling problem. When I insert the tape, it
> > makes sound like it's rewind the tape. Would it be able to do that?
>
> Yes of course it would. The drive is fairly smart, and the first
> thing its going to do when it senses that a tape has been inserted is
> to rewind it, then inspect the header to determine what kind of a
> tape it is, and if it has internal options to drive that legacy tape
> if its say, a dds3 tape being loaded into a dds4 drive. So it will
> make some noises all by itself while doing this.
>
> Now, I hate re-teaching how to re-invent a wheel here, the wheel thats
> called 'scsi cableing, care and feeding' or some such silly attempt
> at being a smart-ass on my part. I have done a nomogram like this
> before, on this list, but not recently. I am a broadcast engineer,
> where transmission line termination errors can easily cost us $50,000
> or more in stuff burned up. But here goes anyway. Bear with me
> folks.
>
> lecture mode on
>
> 1. A scsi cable is a transmission line, and as such absolutely must be
> properly 'terminated' at both ends of the cable, and only at the
> ends. As little as 6" of unused cable hanging from one end or the
> other, past the connectors that are in actual use, is enough to setup
> some echo conditions in the signal transmission that will wreck any
> chance of data integrity being attempted to be sent over the cable.
>
> Therefore, the card must have its terms enabled unless there are
> cables on both the internal connector and the external connector.
> Said another way, if the card has an external connector, but it is
> not in use, then the cards terms must be turned on. This can be in
> older cards, done by plugging in the 3 termination resistor packs in
> the SIL sockets provided on the card, or in later cards, possibly by
> a software option accessable via the cards own bios extensions that
> you see flash up on screen for a few seconds during the machines
> post.
>
> 2. Likewise, the far end termination must be done at the physical end
> of the cable, not at some socket in the middle of the cable. No
> unused cable is allowed to be hanging off the last used connection.
>
> Drives also vary these days in how they handle the termination, with
> far more drives requireing the resistor packs to be installed than
> not, although there is motion toward the use of what we call active
> terminations in this field too. If the drive isn't terminated, its
> not going to work UNLESS its not the last drive on the cable, in
> which case its perfectly legal for it not to be terminated. Its the
> last device that must be terminated.
>
> 2a. These terminations are the only 'loads' on the cable. The drive
> itself, and the cable, are all designed in a wired OR configuration,
> where the terms provide the logic one pullup voltage, and any device
> on the bus can turn on a transistor and pull that data line to
> ground. When they are 'off', the devices present a very small
> loading on the cable so that many devices can share a cable.
>
> There is extensive collision avoidance built into the protocol, so
> that when more than one device does this, its detected and dealt
> with. Silently, transparently, and often with no more than a couple
> of milliseconds delay in completing the data exchange the device
> requested.
>
> 3. There are extant, many cards whose term voltage isolation diodes
> are common power type silicon diodes. Some, and I have no idea if
> the buslogic is among them, have used a schotkey diode for this,
> cutting this voltage loss by about 2/3rds. The reason for the diode
> is to prevent the case where the computer psu is turned off, but an
> external drive enclosure is not, and that drive enclosure is also
> supplying termination power to run these resistor packs to the 'TP'
> line in the cabling. In that event, the computer would then be
> powered from the drive exclosure if it weren't for this diode
> preventing it.
>
> Where the type of the diode becomes a factor is when one is dealing
> with a cable that despite so-called proper termination on both ends,
> is still allowing some ringing on the edges of the signal
> transitions. When the term supply is reduced from its designed value
> of 5 volts because of losses in this diode, then the logic one noise
> margin fades away from its designed value of having the logic one
> resting voltage, wholly established by these terminators, fall from
> the designed value of 600 millivolts above that voltage which is
> guaranteed to be a logic one (2.4 volts), to as little as 100
> millivolts, at which point it doesn't take much ringing to get a dip
> down into the voltage range that is officially defined as
> 'indeterminant', and you have a detected logic error, not often
> correctly reported in the logs, just a coverall, sometimes
> meaningless, error that its not working.
>
> This has become a much smaller consideration where the so-called
> active terms are in use.
>
> But lets talk 'transmission lines' a bit, at the physical level of the
> cable you've had in your hands already.
>
> That cable, when a length of it is inspected with the measureing tools
> available, and considering that every alternate wire in the cable is
> a ground wire (unless the cable is being used in an hvd or lvd
> system), will have a characteristic impedance of about 122 ohms, give
> or take 10 or so due to tolerances in the ribbon cables manufacture.
> Now, if that cable has a 122 ohm load on both ends, a signal
> traveling down its length will be absorbed in these resistors, and
> nothing will come back like an echo, but then superimposed on the new
> voltage level this signal represents. The signals ring in other
> words, if looked at with a sufficiently broadband oscilloscope. A
> 100mhz scope is barely adequate.
>
> To achieve this termination way back in time 35 years ago, and still
> in use today, it was common to use these resistor packs, which
> consist of a 220 ohm resistor with one end tied to the term power
> supply of nominaly 5 volts, the other end connected to the data line
> being terminated, and a 330 ohm resistor is also tied to this data
> line with the far end of it being grounded, those values being used
> because thats what the resistor makers had to offer at Orville
> Redenbacher's "popcorn" prices. This establishes a termination
> resistance value by the normal rules for paralleled resistances of
> around 132 ohms.
>
> Moderately close, and given correct noise margins, a generally
> workable scheme. But, what happens when that 5 volts is reduced by
> the nominally .6 to .75 volt drop of the silicon diode in series with
> that voltage? Well, the expected 3 volts for a logic one can drop,
> often to the point that this resistive scheme so carefully worked out
> 35 years ago, actually presents only a 2.58 volts logic one level,
> and only 180 millivolts of the 600 millivolt noise margin is left.
> Then couple that with the fact that the psu itself may be sagging and
> the 5 volt line is only 4.85 volts at the connectors where the drive
> is drawing power from. Then you have a logic one voltage of only 2.4
> volts, the noise margin has all been used up and no amount of virgins
> sacrificed is going to make it work. Ever.
>
> When the terminations are made 'active', this is marketing speak for
> having a seperate 3 volt regulator set up on the card, and possibly
> the drive, with enough 120 ohm resistors coming off of it to feed
> every data line in the cable by its designed 3 volts, and it comes
> closer to actually matching the impedance of the typical ribbon cable
> to boot. And it has the added advantage of only drawing power when
> the bus is active, unlike the resistor packs, which drew 320
> milliamps from a sometimes scarce power resource when resting, and
> proportionatly more when active, in portable applications this became
> a major source of power loss to be gotten rid of.
>
> In short, active terminations beat passive any day of the week when
> the variations of the real world are plugged into the formula.
>
> In your case, I suspect the drive is not terminated at all, and the
> cable is, electrically speaking, ringing like a bell. Or that you
> have the drive connected at a cable plug that is not the last one on
> the cable. The quality of the terminations can only be assessed with
> high bandwidth oscilloscopes, but a very very good idea can often be
> had just by measuring the resting voltage of a data line with
> everything powered up and that can be done with a $20 dvm from radio
> shack.
>
> If its above 2.9 volts on the cable with first one end unplugged, then
> plug that one back in and unplug the other and recheck, then the
> chances are you are pretty good and have other hardware problems.
> No, or very little voltage when one end of the cable is unplugged
> means the device the cable is still plugged into isn't terminated and
> this must be fixed.
>
> If its only 2.65 to 2.75, its going to be a problem child
> occasionally. Below 2.5 and its likely its not going to work until
> the proper voltage is re-established.
>
> I haven't dealt with the logic zero noise margin considerations
> because its a zero if the line is pulled down to less than .6 volts,
> and I've not seen the scsi bus driver yet that couldn't pull the line
> to well below 25 millivolts, hence its generally not a problem. If
> it can't, the driver chip is toasted, go get someone who knows which
> end of a soldering iron gets hot. I do, but I've yet to have to
> replace one because of that.
>
> Now, I hope I've helped to clarify this thing called a 'scsi bus'.
>
> lecture mode off
>
> > This is my first time setting up a tape drive, it didn't sound like
> > it should be this difficult.
> >Thanks,
> >ERS
> >
> >On Tuesday 15 March 2005 12:14, Eric Shibata wrote:
> > >HELP!
> > >I can't even get the status. No matter what I do I get the
> > > following error.
> > >
> > >/dev/tape: Input/output error
> > >/dev/st0: Input/output error
> > >
> > >Tape drive: HP SureStore T4i
> > >
> > >ERS
> >
> >That sounds a bit like cabling and or termination problems.
>
More information about the fedora-list
mailing list