jlo@elan.elan.com (Jeff Lo) (09/21/89)
We have a problem. We currently have a single thin ethernet running through our office, with a mix of Unix and DOS machines connected to it. All of the Unix machines use IP to talk to each other, and the DOS machines are using something called Network-OS which uses something (?) other than IP. The problem is that we are hitting the maximum length restriction of the thin ethernet and some of the PC's are experiencing difficulties, presumably because their ethernet boards are having problems with the too-long cable. Since the DOS machines don't speak IP, we cannot simply split the ethernet into two separate networks with an IP router in between. So, what I am looking for is something I put on the line, a repeater or something, if such a thing exists for ethernet, to allow us to seemingly have one long (>185m) ethernet. Thanks! -- Jeff Lo, Elan Computer Group, Inc. jlo@elan.com, ..!{ames,uunet}!elan!jlo 888 Villa Street, Third Floor, Mountain View, CA 94041, 415-964-2200
mrichey@orion.cf.uci.edu (Mike Richey) (09/21/89)
In article <634@elan.elan.com> jlo@elan.com (Jeff Lo) writes: >We have a problem. We currently have a single thin ethernet running through >our office, with a mix of Unix and DOS machines connected to it. All of Yea, boy, okay. Have you tried removing devices to see if one of those are causing trouble? There is a limit on the number of _breaks_ that can be in the thin net segment. By beaks I mean connections with T connectors and barrel connectors. If you have one station, that has a T. in the thin net segment there are two _breaks_, one on each side of the T connector. If you have a barrel connector in the segment, there are two _breaks_. Each thin net segment is allowed 60 _breaks_. This translates into 30 stations. This is an impedance parameter of thin net. So if you go beyond the maximum number of _breaks_, whether it be T's,barrels, or patch cords, they all have an affect. Michael S. Richey Internet: mrichey@orion.cf.uci.edu Bitnet: MRichey@UCI CompuServe: 71650,3132 Voice: (714) 856-8374 University of California, Irvine Network and Telecommunication Services 342 Computer Science Irvine, CA 92717
perry@ccssrv.UUCP (Perry Hutchison) (09/21/89)
In article <634@elan.elan.com> jlo@elan.com (Jeff Lo) writes: > what I am looking for is ... a repeater or something, if such a thing exists > for ethernet, to allow us to seemingly have one long (>185m) ethernet. Such a thing does exist. We have here a 3Com box called a MultiConnect Repeater. I believe it can support up to 15 branches, each of which can be of the maximum length permitted for whatever connection technology that branch uses. We currently have two thin-ethernet branches, one of about 150m and the other about 250m. (We are aware that the 250m branch exceeds specification. It nonetheless works pretty well most of the time. We are planning to split it in half -- this will require adding a third interface card to the repeater.) I suspect that 3Com may not be the only manufacturer of such things, and that this particular repeater may not be their only offering. I would suggest checking with (at least): 3Com Western Digital Gateway the manufacturer(s) of your installation's Ethernet cards. No particular recommendations implied here -- I just happen to know that 3Com, WD, and Gateway make Ethernet products. There are almost certainly others.
peiffer@umn-cs.CS.UMN.EDU (Tim J. Peiffer) (09/22/89)
We had a problem here that was interesting, and easy to fix, but I am confused at what I did not see. Why did our repeater not partition under a large impedance mismatch? I expect that the repeater should have seen many collisions... We have segmented our net into several lines where the aggregate lengths are less than 600 ft. In the problem we had, one of our people were installing some portions of a cluster and accidentally installed a 75 ohm at the end. I expected that signals would be broadcasted equally across the back plane to the indvidual repeater modules, and would be echoed back as a collision. Since the mismatch was not really severe, and the aggregate length is ~435ft, was it that the mismatch related reflection was absorbed by the lossyness of the thin-net, and not seen by the repeater? The PC's and the MacII's failed to function as expected, but the repeater failed to partition. I am a bit confused. I really do not think that the repeater is a problem, only my understanding of what did not happen. 3com Multiconnect +---+ | | +-->> 50 ohm terminator | o---| | | +---> 250 ft >-> 3x25ft >-> 18x6ft between MacII >-> PC/AT -> 75 ohm +---+ Tim Peiffer peiffer@cs.umn.edu or Comp Science Dept ...!rutgers!umn-cs!peiffer University of Minn Mpls, MN
edc@excelan.com (Eric Christensen) (09/22/89)
In article <634@elan.elan.com> jlo@elan.com (Jeff Lo) writes: >We have a problem. We currently have a single thin ethernet running through >our office, with a mix of Unix and DOS machines connected to it. All of >the Unix machines use IP to talk to each other, and the DOS machines are >using something called Network-OS which uses something (?) other than IP. >The problem is that we are hitting the maximum length restriction of the >thin ethernet and some of the PC's are experiencing difficulties, >presumably because their ethernet boards are having problems with the >too-long cable. Since the DOS machines don't speak IP, we cannot simply >split the ethernet into two separate networks with an IP router in between. >So, what I am looking for is something I put on the line, a repeater or >something, if such a thing exists for ethernet, to allow us to seemingly >have one long (>185m) ethernet. Thanks! Ethernet repeater are available form a number of vendor. BICC and Cabletron both make excellent units. If traffic levels are high on your net, a bridge may be a better solution. Since a bridge forwards packets at the MAC layer, it's protocol independent and has the advantage of keeping packets between nodes on one side of the repeater off the net on the other side. They're generally a little bit more expensive than a repeater, but are much more useful aas the network grows larger. Again, there are may manufacturers of ethernet bridges, but I've had real good experiance with Retix, NAT, and Cabletron. One warning when choosing a bridge. Pick a vendor and stick with them. There are incompatibilities in the implementation of the spanning tree protocol that the bridges use to talk to each other. While this doesn't usually cause too much trouble (unless your actually building a spanning tree net topology), I have seen one bridge tell another (from a different vendor) to not forward packets one of its interfaces. This can be a problem if your boss is on the segment that keeps getting shut down! :-) +-----------------------------------------------------------------------------+ | Eric Christensen - Sr. System Administrator - Excelan, A Novell Company | | Email: edc@excelan.COM {ames | apple | mtxinu | leadsv }!excelan!edc | +-----------------------------------------------------------------------------+
hedrick@geneva.rutgers.edu (Charles Hedrick) (09/26/89)
The answers people have been giving you are right as far as they go, but have not mentioned something. Most typically large thin-net installations are done in relatively small pieces, using a thin-net multiport repeater to connect them. The problem with thin-net is that you daisy-chain it, so that if somebody disconnects one machine it affects every other machine down the line. It's rather hard to maintain a network that has 100 machines daisy-chained. So the idea is that you daisy-chain small groups of machines that are fairly close together, and then use a multiport repeater to connect the groups. A multiport takes something like 8 segments of thinnet and one of thicknet. If your systems are fairly close together, you may be able to use just a single multiport repeater. If you have a larger installation, you install a thicknet as a "backbone", and then have several multiport repeaters connected by the thicknet backbone. I'm sure you can get a conventional repeater (i.e. a thing that connects two segments) for thin-net, but normally what you find for thin-net are multiport repeaters. Just about all the standard Ethernet vendors make them: We tend to use Cabletron, but DEC certainly makes them (DEMPR), and I'm sure 3Com and all the other Ethernet vendors do as well. At some point you'll want to use a bridge or router rather than just a repeater. Do that when you have enough traffic that you don't really want all the machines on the same network.
pat@hprnd.HP.COM (Pat Thaler) (09/28/89)
HP makes both 802.3 repeaters and bridges. As others have stated, most ThinLAN repeaters are multiported devices. HP's ThinLAN repeater has 4 ThinLAN (10BASE2) ports and an AUI port. Tim J. Peiffer wonders why his repeater didn't partition with an incorrect terminator. Repeaters will partition a port if they see collisions every time they transmit or if a collision persists for an excessive time. It generally takes over 30 consecutive collisions to cause a partition. Partitioning was designed to isolate an unterminated segment and prevent it from taking down the whole network. The repeater transmits to a partitioned port but ignores packets and collisions from that port. Each 802.3 MAU acts as a current source when it transmits. The transmitted signal is a combination of a DC offset and an AC signal. The AC signal carries the data and the DC offset current establishes a DC voltage on the cable. When two or more MAUs (transceivers) transmit at the same time the sum of their currents causes a larger DC voltage. (Actually the offset is negative, but most people find it easier to think in terms of magnitude.) MAUs detect a collision by sensing that the DC voltage has crossed the collision threshold. > Why did our repeater not partition under a large impedance mismatch? I expect > that the repeater should have seen many collisions... We have segmented our > net into several lines where the aggregate lengths are less than 600 ft. In > the problem we had, one of our people were installing some portions of a > cluster and accidentally installed a 75 ohm at the end. > > 3com Multiconnect > +---+ > | | +-->> 50 ohm terminator > | o---| > | | +---> 250 ft >-> 3x25ft >-> 18x6ft between MacII >-> PC/AT -> 75 ohm > +---+ > The larger the resistance the larger the resultant voltage will be. If the resistance gets too large, a collision may be detected when only one MAU is transmitting. The usual causes of excessive resistance are an unterminated cable, an incorrect terminator value, or an excessively long cable (because of the series resistance of the cable). In the case above where a 75 ohm terminator is placed at the far end of the cable, a transmitter near the 75 ohm terminator sees a higher voltage than the repeater's MAU will see when it transmits. It is fairly likely that the repeater's MAU will not see any false collisions even though the PC/AT might see a collision every time it transmits. The repeater therefore will not partition. Pat Thaler
spurgeon@ut-emx.UUCP (Bud Spurgeon) (09/28/89)
In article <16069@umn-cs.CS.UMN.EDU> peiffer@umn-cs.cs.umn.edu (Tim J. Peiffer) writes: >Why did our repeater not partition under a large impedance mismatch? I expect >that the repeater should have seen many collisions... If your wire had been completely unterminated the repeater would have stayed partitioned just fine. But as soon as the repeater can get ONE good packet onto the failing wire, it will stop partitioning. Then after >32 consecutive packet failures it will partition again (but still keep trying to send packets onto the partitioned segment) and if the segment isn't a *complete* lose it will succeed in transmitting a packet onto it and stop partitioning. Depending on how bad your wire mismatch/damage is this can happen quite rapidly and make it appear that the repeater hasn't partiioned the segment at all. As far as the hapless users on the "good" side of the repeater are concerned, it might as well not have tried since the effect is tons of collisions as the repeater jams each failing packet and keeps counting till >32 is reached, then goes back into transmission after one good packet gets through, then starts jamming again, over and over. The moral seems to be that auto-partitioning is good for isolating nice solid wire failures, but if your failure is marginal then the auto-partitioning feature can't really protect you from it.