> On Friday 11 October 2002 12:34, Bernhard Beck wrote: > >FWIW: We had similar symptoms happening here with an external HP > > DLT1 drive off a motherboard-internal Adaptec Ultra2Wide > > controller. The tape was running U2W in single-ended mode and was > > connected using the HP provided LVD/SE cable. It appears the SCSI > > bus was too long with that cable. After switching to a shorter > > external cable (only about 30cm instead of 150cm) all SCSI > > problems/tape errors disappeared. > > > >Bernhard > > That would be pretty prima-faci evidence to me of a boogered > termination someplace. The max length spec for the LVD cabling is > ISTR several times longer than for std single-ended cableing,
I thought of the termination, too. The HBA (AIC7896 Ultra2Wide) is terminating its end, the tape drive has an external LVD/SE termination, all connectors and cables are standard 68pin wide. The tape drive is the only device on this channel. The holding disks are on the second channel of that controller. The only explanation I could come up with to make sense from this behaviour was that the bus (for whatever reason) drops to UltraWide single-ended, which restricts electrical cable length to 1.5m. > which > is by itself about 40 meters! > > By going down to a 30cm cable, the echos are now so close that the > circuitry isn't fast enough and doesn't see them. Don't know that much about the electrical part, but seems to make sense to me as well. Either way, the short external cable fixed or at least avoids the problem for me and the box has been stable over the last few days. Since this is a production box, I can't play around with it as much as I'd like to get this resolved. Maybe on a week-end some time. Bernhard > One other effect that I've no experience with might be that the > LDV/SE cable might be one of the newer miniature cabling styles, > resembling the very finely spaced ata100 cables. Its > characteristic impedance is an unknown to me, whereas the regular, > much wider cable is about 120 ohms. I'd think the smaller cables > characteristic impedance might be a bit lower unless the wire size > shrank in unison with the square laws involved in calculating > impedances. And thats a huge guage shrink... Not practical to do > in an IDC assembly style IMO. >
