derrick@cs.AthabascaU.CA (Derrick Rowlandson) (02/26/90)
Thanks to all who replied, as requeste here is a summary of replies. In article <1676@aurora.AthabascaU.CA> you write: > > > >After spending a couple hours trying to cable up some SCSI peripherals >to a SUN i would like to pose the following questions. > >A SCSI bus consists of a 50 pin bus running from a SCSI controller. >Bus length can be up to 15 feet. >On this bus you can have up to 8 devices. >Connectors for getting on the bus come in many different flavors, >(Centronics style, AMP-socket connectors, D style connector) >Some SCSI peripherals provide a bus extention to allow the bus to >continue on to the next device. This is simply a parallel extention of the bus. > >so far so good? yes. > >Where must a SCSI bus be terminated? Only once, and at the last device on >the bus would seem to be the logical choise. Actually twice - at both ends of the bus (read transmission line). > >When taping into a SCSI bus, the tap or stub lenght can only be a few >inches in lenght i guess? But can you get away with a longer stub lenght >if you also terminate that specific device? NO. You would get nasty reflections no matter what you did. As it turns out SCSI isn't horribly fast so you might get away with it, but random errors may show up. =============================================================================== >Where must a SCSI bus be terminated? Only once, and at the last device on >the bus would seem to be the logical choise. Twice - both ends of the bus. The host doesn't have to be at an end, either, although this is very common. >When taping into a SCSI bus, the tap or stub lenght can only be a few >inches in lenght i guess? But can you get away with a longer stub lenght >if you also terminate that specific device? I've seen stubs have problems at about a foot, and work at about 5 feet. It all depends on the devices in use. Best advice, don't risk your data - do it right. Having an extra terminator is definitely out - will cause lots of problems. Watch out for who is providing terminator power - you can easily blow a fuse (many devices provide fuses expicitly for this) which will cause stange and mysterious problems. =============================================================================== > Where must a SCSI bus be terminated? Only once, and at the last device on > the bus would seem to be the logical choise. No, twice - at each end. But the host adapter is usually at one end, and it is usually terminated at the factory, so you don't need to fool with it. > When taping into a SCSI bus, the tap or stub lenght can only be a few > inches in lenght i guess? But can you get away with a longer stub lenght > if you also terminate that specific device? The standard lists 6.0 meters max, single-ended SCSI bus, with a stub length of 0.2 meters, max, at any device. The reason you can't have a stub of any length that is terminated is that you basically end up with two infinite (looking) lines coming out of the branch, and the signal strength is then halved at the branch. By limiting the length of the stub, and not terminating it, then the reflection comes back "immediately" to add to the rest of the signal going down the main transmission line. (This explanation may have a large bullshit factor. Please post or email a summary of what you get.) =============================================================================== You must terminate an SCSI bus at both ends (first device and last device). No other termination is allowed or necessary. The 15-foot length is for single-ended SCSI only; differential allows longer. =============================================================================== so far not so good. maximum spec-ed length is 6m. the bus passes through devices. no parallel connections are allowed although i think a couple of short stubs within a cabinet is within the spec. =============================================================================== > Bus lenght can be up to 15 feet. I believe the spec allows 6 meters for nondifferential SCSI, and 25 meters for differential SCSI. > On this bus you can have up to 8 devices. But your host adapter counts as one of the 8. > Where must a SCSI bus be terminated? Only once, and at the last device on > the bus would seem to be the logical choise. Close. I believe it is supposed to be terminated on each endpoint of the cable. Since one endpoint is likely to be inside your host, it may already be terminated. > When taping into a SCSI bus, the tap or stub lenght can only be a few > inches in lenght i guess? But can you get away with a longer stub lenght > if you also terminate that specific device? My guess is that long taps are not allowed, even if you terminate the taps. Most SCSI boxes I've seen go to the trouble of routing the cable in to the SCSI devices, then back to the outbound connector. Adaptec used to print a good book on SCSI. Maybe they still sell them or give them away. It has been 4-5 years since I got my copy. =============================================================================== My books say 18 ft. is okay. Only the last device may be terminated. =============================================================================== SCSI host adapters typically terminate their end internally. You have to worry about the other end. The official length limit for "single-ended" SCSI (this refers to the bus drivers, the alternative is "differential") is 6 meters. The maximum stub length is 3 inches. You can usually get away with a lot of cheating if your cables are short. Missing termination at one end or an extra termination in the middle is often not fatal. Slightly longer stubs are probably ok. If you must cheat, make sure SCSI bus parity checking is enabled everywhere. =============================================================================== Your assumptions are pretty much on the money. SCSI busses are to be terminated at each end of the Bus. SCSI host adapters (Your 3/50 for example, or a VME SCSI card or whatever) have terminators built in. You should also terminate the last device. As for the stubbs, if you look inside a SCSI shoebox (with both in and out connections) you will see that there is a ribbon cable thet looks from one to the other. An inline 50 connector is cripmed onto the middle of the cable, for each device in the showbox, and plugs into the device. There is no stub except for the SCSI device PCB itself. .reality on In reality you can abuse things quite badly and SCSI will still work. It is very forgiving. You can usually leave one of the terminators off and not get burnt. I have also seen really scary cabling setup where they built a star type cable out of ribbon cables. It worked, but I would hate to depend on it. The reflections must be terrible. If you are having difficulty seeing a device try another cable and make sure things are addressed properly. Unless you're doing something really ridiculous your cabling scheme is likely not the problem. .reality off =========================================================================== 1. The maximum length of a single-ended (non-differential) SCSI bus is 6 Meters. 2. Termination must be at both ends of the bus, and nowhere else. Typically, the bus is terminated at the HBA on on end, and at the last device or with an external terminator at the other end. 3. Maximum stub length is 0.1 Meters. 4. Maximum number of devices on a bus is 8, incuding the Host Bus Adaptor (HBA). 5. Many people violate some of the above, and the bus still works. You will find less chance of a non-conformant bus working as you add devices and increase the speed (i.e. synchronous operation). Some devices are more sensitive to bus variations than others. =========================================================================== Twice... once at each end. If your host (PC, Mac, whatever) is at one end of the bus, it should have a terminator. If you have a very short bus (e.g. a Mac and a single hard disk connected via a 1' cable), a single terminator at the disk may suffice. > When taping into a SCSI bus, the tap or stub lenght can only be a few > inches in lenght i guess? The SCSI-1 spec says ".1 meter", or roughly four inches. > But can you get away with a longer stub lenght > if you also terminate that specific device? Adding a third terminator to the bus will probably cause it to cease functioning. The terminators do more than terminate the bus with the correct impedence... they also act as pull-up circuitry, holding the bus (in its idle state) at roughly 3 volts. If you have too many terminators on the bus, a SCSI device may not not be able to sink enough current to pull the bus down to the low-voltage state... and it may thus be unable to transmit properly. You may be able to get away with a longer stub if your bus is not heavily populated, and/or your SCSI devices are robust, and/or you're not pushing the bus to its limits with high-speed synchronous data transfers. You may be able to _eliminate_ a stub entirely. If your SCSI box has a pair of connectors ("SCSI IN" and "SCSI OUT", for example), you can wire the jacks together with a horseshoe-shaped loop of ribbon cable, long enough to reach over to the SCSI device(s) in the box. This way, the loop inside the box becomes part of the SCSI bus, rather than being a stub. > confirmation/correction of the above would be greatly appreciatted, and > any other cabling/hardware tips also welcome. Hacking SCSI setups is a bit of a black art. There's a lot of science to it... but it seems that the cut&try method is still necessary a lot of the time. =========================================================================== UPTO 20 feet and that includes the bus lenght used *inside* the enclosure of choice. >On this bus you can have up to 8 devices. Actually only 7. The 8th (target 7) is the host adapter itself. >Connectors for getting on the bus come in many different flavors, There are only two that are actually true standard: the champ or Centronics style, and the DB-50 style. Sun's new tiny thing on the SPARCStations is their own at this point. >Some SCSI peripherals provide a bus extention to allow the bus to >continue on to the next device. This is simply a parallel extention of the bus. This is actually a continuation of the bus. >Where must a SCSI bus be terminated? Only once, and at the last device on >the bus would seem to be the logical choise. Yes. It is the last physical device not the highest id's device. >When taping into a SCSI bus, the tap or stub lenght can only be a few >inches in lenght i guess? (LENGTH) This stub can only be 5inches long. >But can you get away with a longer stub lenght (LENGTH) >if you also terminate that specific device? No. 1. If you want your system to run reliabably don't exceed the 20foot limit. 2. Besure all your connections are screwed down. 3. Do not bend those thick cables too tight or you'll pull the pins out. 4. Don't connect or reconnect devices while power is on. Failure to follow the above points is asking for hardware failure and/or scrambling the image on the disk. =========================================================================== >A SCSI bus consists of a 50 pin bus running from a SCSI controller. yes, though there is a more-or-less standard 25-pin version (25+ of the normal 50 are ground). >Bus lenght can be up to 15 feet. Actually 6 meters. >On this bus you can have up to 8 devices. That includes the host, remember. >Some SCSI peripherals provide a bus extention to allow the bus to >continue on to the next device. This is simply a parallel extention of the bus. Sort of. There are sub-addresses which can be dealt with in any way the device wants to (Logical Unit Numbers (LUNs), ranging from 0-7). Most devices implement only one LUN (0). >Where must a SCSI bus be terminated? Only once, and at the last device on >the bus would seem to be the logical choise. Normally yes (though I think the host adapter may need termination, I'm a software guy so I'm not sure without checking). >When taping into a SCSI bus, the tap or stub lenght can only be a few >inches in lenght i guess? But can you get away with a longer stub lenght >if you also terminate that specific device? ========================================================================= No. A SCSI bus must be terminated at each end of the cable. If the cable starts inside a computer and, via an external connector, exits to the outside world and connects to a device, then the terminators must be removed off the adapter and the bus would be terminated at both ends of the cable. >When taping into a SCSI bus, the tap or stub lenght can only be a few >inches in lenght i guess? But can you get away with a longer stub lenght >if you also terminate that specific device? A stub is not a good idea. This leaves you open to all kinds of unreliable operations. If you have a controller that supports a number of LUN's, then termination of those LUN's are handled independent of the SCSI bus. The controller itself is the only main concern. =========================================================================
kc@hprnd.HP.COM (Kurt Chan) (03/01/90)
> The standard lists 6.0 meters max, single-ended SCSI bus, with a stub length > of 0.2 meters, max, at any device. Only .1 meters for single-ended. The .2 meter stub length is when differential transceivers are used. > By limiting the length of the stub, and not terminating it, then the > reflection comes back "immediately" to add to the rest of the signal going > down the main transmission line. (This explanation may have a large bullshit > factor. Please post or email a summary of what you get.) No BS - this is essentially correct. A rising signal being propagated down the bus will indeed be attenuated at device stubs, in proportion to the reflection coefficient produced by the stub in parallel with the main bus. An 80 ohm main bus in parallel with an 80 ohm stub will appear as a 40 ohm discontinuity in the cable. Some of the initial signal will be reflected back to the source, and another portion will continue on down the main bus. The *magnitude* of the "glitch" is proportional to the relative impedances. The *duration* of the glitch is a function of the stub length. The longer the length, the more likely a device will see a low-going pulse where none was intended. > My guess is that long taps are not allowed, even if you terminate the taps. Please don't "terminate" these taps! As noted, only two termination circuits are permitted, one at each end of the bus. Adding termination circuits may damage the drivers on any of the attached devices. It may also degrade the low-level noise margin by increasing V(ol) beyond the 0.5V maximum. > Most SCSI boxes I've seen go to the trouble of routing the cable > in to the SCSI devices, then back to the outbound connector. Short of using special "Y" cables, having two connectors is the only way to daisy chain the cables used on drives with shielded bulkheads. > SCSI host adapters typically terminate their end internally. You have to > worry about the other end. In general, but not always. For example, some host adapters are actually in the "middle" of the bus - peripherals internal to the host cabinet are at the end. The connector you see may have to be terminated if there are internal but no external peripherals, or if there are external but no internal peripherals. If there are both, the host adapter would have to be "unterminated". > You can usually get away with a lot of cheating if your cables are > short. Missing termination at one end or an extra termination in the > middle is often not fatal. Slightly longer stubs are probably ok. > If you must cheat, make sure SCSI bus parity checking is enabled > everywhere. In general, the bus doesn't become a transmission line until the propagation delay exceeds the rise/fall time of the signals it carries. At transition times of 5ns and cable speeds of about 1.5-2ns/ft, this means the cable had better be no greater than about 24". > In reality you can abuse things quite badly and SCSI will still work. It > is very forgiving. I've found the opposite to be true, when using single-ended systems with more than 4 devices on cable lengths longer than 4 meters. > You can usually leave one of the terminators off and not get burnt. But please don't try it! > Unless you're doing something really ridiculous your cabling scheme is > likely not the problem. I've found cabling and termination to be the primary source of problems on maximally configured systems. Use high-impedance cables, and Alternative-2 termination! > UPTO 20 feet and that includes the bus lenght used *inside* the > enclosure of choice. Good point - some devices use up to a meter or two for internal cable routing. >>Connectors for getting on the bus come in many different flavors, > There are only two that are actually true standard: the champ > or Centronics style, and the DB-50 style. Sun's new tiny thing on > the SPARCStations is their own at this point. There also are new "high-density" connectors with side latches instead of bail lock clips defined in SCSI-2. They are much smaller than the Champ connector and may be what you're seeing on Sun's products. > (LENGTH) This stub can only be 5inches long. 0.1 meters = 3.937"