phil@amdcad.UUCP (Phil Ngai) (07/09/86)
In article <145@ima.UUCP> johnl@ima.UUCP (John R. Levine) writes: >In article <218@cfa.UUCP> wyatt@cfa.UUCP (Bill Wyatt) writes: >>Remember that a DELNI takes the place of a transceiver. Transceivers do >>not pass carrier sense through to an interface. Some of those ten wires in a >>transceiver cable are input, others are output. See my original article on >>the difference between an isolated DELNI and a connected one. > >Wrong. Incorrect. Untrue. (Sorry to be rude, but it dismays me to see >people firmly explaining their misconceptions.) The DELNI is not a >transceiver, it is a transceiver multiplexer. All of the stations >downstream from a DELNI need transceivers of their own. A connected DELNI >acts like a repeater from the downstream stations to the upstream cable, and >I presume nobody denies the existence of Ethernet repeaters. I am very surprised to see John Levine passing around such misinformation. I had come to expect much higher quality from him. Let me quote from EK-DELNI-TM-001, "DELNI Ethernet Local Network Interconnect Technical Manual": "1.1.1 Communication Interfaces The DELNI unit provides nine communication interface. These are: o Eight (8) local connectors (labeled 1 through 8) and o One (1) global connector (labeled with a square) 2.1.3 Connected DELNI LAN In a Connected DELNI LAN application (Figure 2-3), the stations are connected to the local connectors of a single DELNI unit operating in the GLOBAL mode. The global connector of the DELNI unit is connected to an Ethernet transceiver." Figure 2-3 confirms what should be obvious from this description, that the stations are directly connected to the DELNI without transceivers. The DELNI itself uses a transceiver to access the main Ethernet coax. Also, the DELNI does not act like a repeater in the Ethernet sense of the word. It does not retime or regenerate preamble. >The real problem with cascaded DELNIs is that the DELNI introduces no >delay to a packet that's passed through it, but it takes a moment to >realize when a packet has started, so a little bit of the packet preamble >is nibbled off. After several levels of DELNI, you lose enough of the >preamble that some stations can't recognize the packet. The specs are that the squelch (both TRANSMIT and RECEIVE) is from 100 to 250 nS and the propagation delay is from 10 to 25 nS. This means that an extra level of DELNI cascading would eat as much as 5 bits from the 64 bit preamble. This assumes an extra level up to the coax and an extra level down from the coax but that only transmitter squelch applies on the way up and receiver squelch on the way down. 250 plus 250 equals 500 nS or 5 bits. Compare this to the requirements on the MAU imposed by 802.3 (Ethernet 2 is identical): 8.2.1.1 Transmit Function Requirements At the start of a frame transmission on the coaxial cable, no more than 2 bits of information may be received from the DO circuit (transmit circuit of the transceiver cable) and not transmitted onto the coaxial medium. In addition it is permissible for the first bit sent to contain invalid data. 8.2.1.2 Receive Function Requirements At the start of a frame transmission from the coaxial cable, no more than 5 bits of information may be received from the coaxial cable and not transmitted onto the DI circuit (receive circuit of the transceiver cable). In addition it is permissible for the first bit sent over the receive circuit to contain invalid data. So we have a total of 3 plus 6 bits gobbled by the transceiver/MAU path (one into and one out of the coaxial cable). Section 7.5.1.2 of the Ethernet 2 spec requires the decoder of the station to operate with only 16 bits. I don't know if 802.3 has a similar requirement. For a non-DELNI system, there is 64 - 9 - 16 = 39 bits of preamble margin. For a single level DELNI system, there is 64 - 9 - 16 - 5 = 34 bits of margin. For a two level DELNI system (illegal by DEC configuration rules) there is 29 bits of margin. (fudge the above numbers by a couple of bits to allow for seeing the 11 which ends the preamble) I may have missed something here but it doesn't seem to me that truncated preamble is a reason for limiting Connected DELNIs to one level. Despite having just argued to the effect that a cascaded Connected DELNI should work, I would not recommend the use of such a configuration. There may be other reasons DEC does not support such a configuration. >I have personally lashed up 3-deep connected DELNIs and it worked, but I >was lucky because all of the transceivers involved could deal with the >truncated preamble. Actually, it's the station that has to deal with truncated preamble since it has to generate a clock out of the incoming bit stream. Yes, it may work at the time when you install it but the next time you expand the network, will it still work? I must also confess to being confused by Bill's statement that "transceivers do not pass carrier sense through to an interface". Assuming he means a station when he says "interface", is not the presence of signal on Circuit DI (received data) of the AUI cable an indication of carrier? And what are the implications of this with regard to cascaded DELNIs? -- Bring back The Phone Company! Phil Ngai +1 408 749 5720 UUCP: {ucbvax,decwrl,ihnp4,allegra}!amdcad!phil ARPA: amdcad!phil@decwrl.dec.com
wyatt@cfa.UUCP (Bill Wyatt) (07/09/86)
> > I may have missed something here but it doesn't seem to me that > truncated preamble is a reason for limiting Connected DELNIs to one > level. > > Despite having just argued to the effect that a cascaded Connected > DELNI should work, I would not recommend the use of such a > configuration. There may be other reasons DEC does not support such a > configuration. > >>I have personally lashed up 3-deep connected DELNIs and it worked, but I >>was lucky because all of the transceivers involved could deal with the >>truncated preamble. > > Yes, it may work at the time when you install it but the next time you > expand the network, will it still work? > > I must also confess to being confused by Bill's statement that > "transceivers do not pass carrier sense through to an interface". > Assuming he means a station when he says "interface", is not the > presence of signal on Circuit DI (received data) of the AUI cable an > indication of carrier? And what are the implications of this with > regard to cascaded DELNIs? All right! I give up. As I mentioned in the original posting, I was passing along info from someone who was not himself an expert on the technical aspects of Ethernet, but had had it explained to him. Obviously, this third-hand info is not ideal, and I don't know how what the real problem is. One thing that was definite: cascaded connected DELNIs are unstable- that is, they may work and then later cause massive problems when a network expands. Isn't there anyone out there, preferable the design engineer at DEC, who *really* has the straight story? -- Bill UUCP: {seismo|ihnp4}!harvard!talcott!cfa!wyatt Wyatt ARPA: wyatt%cfa.UUCP@harvard.HARVARD.EDU
bartlett@encore.UUCP (John Bartlett) (07/10/86)
There are a number of parameters that I am sure DEC is considering
when they disallow cascaded DELNI, and the comments from Phil on
lost preamble bits are one of the issues.
The thing Phil did not address in his calculation of preamble bits
is the presense of other bit eaters in the ethernet network such as
repeaters. In a less than full sized network, preamble eating is
not a problem.
The second issue is the total round trip delay time of the network.
Each DELNI does add propagation time delay to the network, and the
total amount of network delay should not be violated because it is
tied to the slot time and used in the CSMA/CD protocol. In any
specific application this may not be an issue, because again a full
sized network may not be implemented. Remember, a full sized
network (from a delay point of view) has as much as 2.5 KM of length
including a 1KM repeater link. If you are operating with a single
500M segment, delay won't be much of an issue.
The last issue is signal jitter. Jitter defines how far in or out
of a bit window a signal can transition, and still be properly
detected by the receiving logic. Each thing (active or passive) in
the network adds jitter to the signals (including the DELNI) and
degrades the signal quality.
DEC recommends not cascading DELNI because they have to insure your
system will work in any configuration. It would be very difficult
to specify how to figure out if you can have x levels of cascading
for a y length network etc. etc.
The Encore Annex has a little single line DELNI that allows one
other station to use the same transceiver that it is using. The way
I specified the network configuration issue for the Annex was to
coorelate the Annex to a certain amount of transceiver cable. In
other words, the unit cascaded through the Annex cannot be more than
50 meters from the transceiver (Ethernet spec) and the Annex itself
looks like 4 meters of cable. Hence the combined transceiver cables
between the Annex and the Transceiver, and the second unit and the
Annex cannot exceed (50 -4) 46 meters.
John Bartlett {ihnp4,decvax,allegra,linus}!encore!bartlett
Encore Computer Corp.
257 Ceder Hill Street
Marlboro, Mass. 01752
(617) 460-0500
Opinions are not necessarily those of Encore Computer Corp.phil@amdcad.UUCP (07/12/86)
Well, I hate to disagree with anyone who makes a terminal server as nice as Encore's Annex but I do have to inject some references into this discussion. In article <320@encore.UUCP> bartlett@encore.UUCP (John Bartlett) writes: > >There are a number of parameters that I am sure DEC is considering >when they disallow cascaded DELNI, and the comments from Phil on >lost preamble bits are one of the issues. > >The thing Phil did not address in his calculation of preamble bits >is the presense of other bit eaters in the ethernet network such as >repeaters. In a less than full sized network, preamble eating is >not a problem. Ethernet Version 2.0, November, 1982, section 7.6.4.1, page 83: "The repeater must transmit 64 preamble bits onto the outgoing segment" ANSI/IEEE Std 802.3-1985, section 9.1.2.3, page 126: "PREAMBLE INSERTION. The repeater unit shall output at least 56 bits of preamble followed by the start frame delimiter." (IEEE decided to be byte oriented and broke up the Ethernet 64 bit preamble into a 56 bit "preamble" and an 8 bit SFD. When you string all these bits together, the result looks just like a 64 bit Ethernet preamble.) >The second issue is the total round trip delay time of the network. >Each DELNI does add propagation time delay to the network, and the >total amount of network delay should not be violated because it is >tied to the slot time and used in the CSMA/CD protocol. In any >specific application this may not be an issue, because again a full >sized network may not be implemented. Remember, a full sized >network (from a delay point of view) has as much as 2.5 KM of length >including a 1KM repeater link. If you are operating with a single >500M segment, delay won't be much of an issue. A DELNI has a delay of 25 nS each way, maximum. (reference EK-DELNI-TM-001 "DELNI Ethernet Local Network Interconnect Technical Manual".) 1000 meters of point to point repeater link has a delay of 5,000 nS each way. 1500 meters of coax has a delay of 6,500 nS each way. This is straining at gnats and swallowing camels if you think the propagation delay of a DELNI is an issue. Well, if you had 460 DELNIs in series you might want to worry. I don't generally deal with things less than 1%. >The last issue is signal jitter. Jitter defines how far in or out >of a bit window a signal can transition, and still be properly >detected by the receiving logic. Each thing (active or passive) in >the network adds jitter to the signals (including the DELNI) and >degrades the signal quality. This could be a reason. The DELNI has 1 to 2 nS of timing asymmetry. The entire system jitter budget is only 18 nS, including margin, in 802.3. Of that 18 nS, 16.5 nS is already allocated. So the use of a DELNI or two already puts the system on shaky ground with regard to jitter margin. Cascading them could very well be a problem because of this. By way of comparison, the MAU is allowed 2 nS in the transmit direction and 1 nS in the receive direction. -- Why did VD become STDs? Phil Ngai +1 408 749 5720 UUCP: {ucbvax,decwrl,ihnp4,allegra}!amdcad!phil ARPA: amdcad!phil@decwrl.dec.com