This is part of The Pile, a partial archive of some open source mailing lists and newsgroups.
Subject: Re: Acceptable number of collisions? From: "Jose M. Sanchez" <opjose@ex-pressnet.com> Date: Thu, 27 May 1999 00:10:59 -0400 Paul Mezzanini <paul.mezzanini@saintleo.edu> wrote: > All a hub does is provide a means to connect a bunch of wires together > so they 'look' like one. (ya know what I mean, Tx to Rx yada yada yada) > > Of course I could be wrong, but I don't think I am. :) Eh, no and yes... Normally moving from 10Base2 to 10BaseT results in a quadrupling of throughput... This is because the TX/RX pairs are on separate lines... With 10Base2 everything is broadcast over one wire and is much slower... Hubs also don't merely connect a bunch of wires together... IBM Token ring hubs DO, ethernet somehow picked up the Tokenring/Arcnet nomenclature which has resulted in a lot of confusion... Ethernet hubs are a misnomer... they are actually called "concentrators" or "signal conditioners" since they receive the signal and "clean it up" again... Switches go one step further by isolating pathways between connected cable segments, since there is less collision retransmissions and often (in better switches) machines can communicate with counterparts on dissimilar segments AT THE SAME TIME, the apparent xfer rate improves... In actuality two machines on a switch are no faster than two machines on a hub... add more machines and the situation changes dramatically if they are all talking to "other" devices... If they are all hitting the same server at the same time, however, performance is not greatly improved... === Subject: Re: Acceptable number of collisions? (NOT SO!) From: "Jose M. Sanchez" <opjose@ex-pressnet.com> Date: Tue, 1 Jun 1999 18:01:53 -0400 Richard W. Gowen <richard@gowen.net> wrote: > Jose M. Sanchez <opjose@ex-pressnet.com> wrote: > > Sorry, NOT SO... > > > > The discussion was one about "collisions". > > In a 10Base2 versus a 10baseT network actual data > > transfer rates are unaffected... that is you do get a > > maximum of 10megabits UNIDIRECTIONALLY going from point > > to point under IDEAL conditions... > That was my point. The physical protocol CSMA/CD > establishes a physical collision domain in both 10baseT > and 10base2. The operations of this domain are the same > for both media types. See:"Local Area Networks 3rd > Edition" by John Slone(c) 1997, Auerbach, Boston MA > > However in a 10base2 network, data is broadcast in a > > manner identical to standard RF transmissions. That is > > the center core of the co-ax cable acts as an antenna... > Excuse me, but I believe that the CSMA/CD protocol uses > Manchester encoding. How exactly is this stepping of > electrical voltage on a wire like an antenna? Manchester encoding on a carrier frequency... The carrier is "broadcast" to all stations over the wire. The electrical signals exhibit the same propagation characteristics as a typical antenna segment... Hence the need for terminators in a 10base2 wire ... > Since you want to keep this discussion to collisions why > are we talking about acknowledgments? Because it is a factor contributing to the number of collisions occuring in a 10base2 network. > design, implementation, and support experience are enough > to make these statements. Apparently not, or you'd be fully aware of the difference data exchange speed between a 10base2 and a 10baseT lan even though the underlying data speed is the same... This is not my hearsay, rather what people have experienced for many years when switching over from 10base2 to 10baseT... > A token ring MAU (multistation access unit) does no signal > conditioning and is nothing more then a passive relay. Hmm. I mentioned this originally and you must have missed this.. > A 10baseT HUB is a bus topology, that is each station's > CAT3 or better cable is physically wired to a central > backing. Not so, -ALL- 10baseT hubs have active conditioning circuitry... again show me one that doesn't. The spec mandates this. > And as you can see, there is no mention of signal > conditioning in the basic requirements of a hub. If you > want a real life example of this I can point you to IBM's > Model 8228 Token-ring MAU. I don't have an example handy > of an ethernet passive hub. You are grouping token ring into a discussion of ethernet hubs, then calling the ethernet requirements superfluous... Token ring is a different beast altogether... > However, I can find one if the issue is pressed. This I'd like to see. > > > Yes you can make a cable... and because of signal degradation and > > propagation problems it will not exceed fifty feet... Give it a try > > sometime! (Better yet, add the third machine in using nothing but > > cables... let's see what happens then...) > > Works fine with four. No. The very nature the design precludes this. > > However for multiple connections the IEEE spec DOES > > require signal conditioning... > Where? I have looked at the CSMA/CD specs. I don't see > what you are referring to. Please enlighten me (RFC > number please). When I get a chance I'll find them for you... > > I can make an ethernet or token-ring hub with none of > > this. In fact you can simply make a crossover cable > > (cross tx and rx wires) to connect two 10baseT machines. > > This is possible because all that is required at the > > physical layer is a connection of wires between the two > > machines. The configuration of those wires varies based > > on topology type (ethernet, token-ring, etc...). > > --- > > Irrelevant, we were talking about throughput vis-a-vis > > collisions on an ethernet lan.... > It is relevant since we are talking about the physical > layer. The collision domain can be set up logically in a > smart hub but only needs to be set up physically. That > was my point. Eh no, your point was that there was no difference when moving from 10base2 to 10baseT... The actual mechanics are not in question... === Subject: Re: [OT]Electronics (Was: Acceptable number of collisions?) From: Tom Burke <tomii@erols.com> Date: Tue, 01 Jun 1999 18:37:38 -0400 > John Summerfield wrote: > > This is complete and utter rubbish. Thicker wire gives > > lower resistance and so allows a greater current flow > > for a given potential difference. [...] Of course, > > this is only true for low frequencies and DC[0], due to > > the skin effect... SCNR, Thomas [0] Ok, for you RF > > folks 10MHz probably *is* DC... ;-) OK, OK.... I gotta pipe up on this one - and it will be long winded :) All electromagnetic waves propagate at essentially the same speed (Speed of Light - c - 300000 meters/second) (Although electrons move considerably slower - imagine a tube full of marbles). This being the case, there must be something different going on with the change in speeds... Here it is: 10Mbit ethernet has a data transfer to 10Mbps (Ideally) during transmit/receive. However, understand that we are talking about PACKETIZED data here. To send a 100MB file at 10Mbps would ideally take 1.33 minutes. This is an unacceptable time for one machine to usurp _ALL_ the network resources. Instead, we break the file up into packets. Depending on the protocol, the packets can vary in size (even within the same protocol). For ease of use, most have single-sized packets. Now, most forms of ethernet rely on a variation of the old ALOHA protocol - everything listens and transmits on the same piece of wire. Think of a dark room with 10 people in it. If you want to talk, you listen & make sure that no one else is talking. Then you pipe up & say whatever you want to say. If you talk at the same time as someone else, you both hear something different from what you said. You stop, wait a random amount of time, then try to say it again. The problem here, is that you can transmit quite a bit of data before the guy who's 100 feet away's data reaches you, so that you know your packet has had a collision (It takes time for that electromagnetic wav to travel). Now, there's some variations on this, & I believe 802.3 uses a marker. If you want to talk, you send a reservation signal down the wire. If you didn't transmit the marker, you must wait a minimum amount of time before you can try to reserve "air time." As the talker, you must wait a minimum amount of time to ensure that al other stations have received your reservation. If you hear a reservation that is not yours (say, two stations at opposite ends tried to reserve at the same time), both wait a random time & try again. This maximum time is defined by the length of the cable allowed to connect the stations. in 10base2, that length is 200 meters. In 10base5, that distance is 500 meters. This distance is also the limiting factor on the speed of transfer, as it limits the number of packets that may be sent from a single host collision free. Now, it seems that if you boost your signalling speed up to 100Mb from 10Mb, you should be able to move 10x the data (on a normal network, with multiple hosts all wanting to talk)... Statistically, this is not the case. In fact, as I recall, it is actually more than that. It is all based in nasty statistics & probability - I'll admit, I'm no good at that stuff :( As far as the 10/100Mb goes, that is the data transfer rate (maximum). I doubt that this is baseband in very many circumstances. In fact, this is probably modulated onto an RF carrier. By Nyquist's sampling theorem, this carrier must have a _MINIMUM_ of 2x the data rate to avoid aliasing of data. I would guess that the carrier frequency is more along the lines of 4 to 5 times the intended data rate. Of course, this all assumes ideal lines, perfect terminations, etc :) For more info, check out: Data Netwrks by Dimitri Bertsekas/Robert Gallager Digital Signal Processing by Alan V. Oppenheim/Ronald W. Schafer Modern Digital and Analog Communications Systems by B.P Lathi === Subject: Re: Acceptable number of collisions? (NOT SO!) From: "Jose M. Sanchez" <opjose@ex-pressnet.com> Date: Tue, 1 Jun 1999 01:55:38 -0400 Sorry, NOT SO... The discussion was one about "collisions". In a 10Base2 versus a 10baseT network actual data transfer rates are unaffected... that is you do get a maximum of 10megabits UNIDIRECTIONALLY going from point to point under IDEAL conditions... However in a 10base2 network, data is broadcast in a manner identical to standard RF transmissions. That is the center core of the co-ax cable acts as an antenna... When transmitting data from point a to point b without any other interfering machine on a lan, due to the lack of collisions on the network data is transmitted as quickly as it is on a 10baseT network PROVIDED that NO acknowledgements are needed... Now TCP/IP sends packets of acks, as do other protocols, so the sender gets interrupted whenever the recipient attempts to ack the data stream... the greatly reduces throughput... this introduces a point of degradation... When more than one machine is present on the lan, even at idle, packets are being exchanged, resulting in further "backing down" due to collisions... Compared this to the same situation on 10baseT. 10BaseT has separate send/receive pairs. In a normal 10baseT network full duplex operation is enabled. The point to point transfer acks no longer cause collisions, since they occur on the other wire pairs) resulting in data streaming at closer to the maximum spec of the lan itself (namely 10mbit/sec)... Toss a few more machines in to the fray, even without a switching hub, because "idle" packets spend less time in collisions avoidance/detection, there is a corresponding decrease in traffic which is interfering with the point to point data transfers... The end result, is that from the users perspective thruput increases greatly when moving from 10base2 to 10BaseT. --- "I think a basic electronics course is in order." --- Eh, I think a real world experience course is due... but hey don't believe me... See:"Analyzing Novell Networks" by Carl Malamud (c) 1990, Van Nostrand Reinhold, New York... Plus many other books on the same subject... And... if their is anyone on the list who has moved from 10Base2 to 10BaseT, using hubs, please chime in on your results please... heh. --- And FYI, coax cable can actually carry more throughput then telephone cable. This is because it has a greater diameter and more area for the electrons to flow across then all four of the telephone wires you use in 10baseT. How do you think cableTV pushes all that data to your television over 2 wires? --- To paraphrase: "Duh..." We were talking about 10baseT versus 10BaseT, not the merits of co-ax over telephone media and band width on both, your statement is irrelevant to the original discussion... which was 10base2 over 10BaseT. But yes co-ax can carry more signal bandwidth than plain old telephone wire, which is why I'm using a cable modem... however how does this affect my original assertion? "go with 10baseT on your home lan" -- "Hubs also don't merely connect a bunch of wires together... IBM Token ring hubs DO, ethernet somehow picked up the Tokenring/Arcnet nomenclature which has resulted in a lot of confusion... Ethernet hubs are a misnomer... they are actually called "concentrators" or "signal conditioners" since they receive the signal and "clean it up" -- again... This misleading if not untrue. -- Hog wash, all hubs are signal conditioners of some sort... point me to one that isn't... Yes you can make a cable... and because of signal degradation and propagation problems it will not exceed fifty feet... Give it a try sometime! (Better yet, add the third machine in using nothing but cables... let's see what happens then...) What part of the above is misleading or untrue? Let's see "Hubs don't merely connect wires together." True, they minimally condition the signal... "Ethernet hubs are signal conditioners." True. -- While most modern HUBS (ethernet and token-ring) do more then just connect wires together. -- Thanks for admitting that, but aren't you implying that your OWN statement is untrue? -- things like signal degradation correction or rmon statistic gathering or snmp trap event reporting, none of this is required. -- You are right, for a point to point connection none of this is required, because you are effectively taking advantage of designed in tolerances... However for multiple connections the IEEE spec DOES require signal conditioning... But this has little to do with your assertion or the original discussion... -- I can make an ethernet or token-ring hub with none of this. In fact you can simply make a crossover cable (cross tx and rx wires) to connect two 10baseT machines. This is possible because all that is required at the physical layer is a connection of wires between the two machines. The configuration of those wires varies based on topology type (ethernet, token-ring, etc...). -- Irrelevant, we were talking about throughput vis-a-vis collisions on an ethernet lan.... -- While the term switch is widely overused in the industry, it usually refers to a segment of a particular topology that is being bridged or "switched" to other segments of the same or different topology. Some of the newer switch manufactures even offer a proprietary backplane between the switched segments that offers high bandwidth. Usually switched ethernet segments support one device per switched segment. This is because with only one device on the segment you have no collisions. This alone gives you better throughput. -- I can't argue with that, this is a statement of fact... which is why I originally told the person go with ethernet... 10baseT over 10Base2 (the original discussion) -- Well, my rant is done. I just can't stand when incorrect information is given out. -- I guess I can't stand it either hence this message... Next time be sure of what you are saying before you assume that someone else is wrong... Better yet, read the WHOLE thread that way -you- will not be the one giving out "incorrect" information again... --- And, yes, I am a network engineer -- heh, I just might be a "little" bit more... and at it a "little" bit longer... === Subject: Re: Acceptable number of collisions? From: John Summerfield <summer@OS2.ami.com.au> Date: Tue, 01 Jun 1999 10:17:11 +0800 > Normally moving from 10Base2 to 10BaseT results in a > quadrupling of throughput... This is because the TX/RX > pairs are on separate lines... With 10Base2 everything is > broadcast over one wire and is much = slower... This is > untrue. The throughput is a constant 10 Megabits. =20 This > is why it is called "10"base<MediaType>. And FYI, coax > cable can actually carry more throughput then telephone > cable. This is = because it has a greater diameter and > more area for the electrons to flow across = then all four > of the telephone wires you use in 10baseT. How do you > think = cableTV pushes all that data to your television > over 2 wires? I think a basic = electronics course is in > order. This is complete and utter rubbish. Thicker wire gives lower resistance and so allows a greater current flow for a given potential difference. It absolutely does NOT allow one to feed more information across it. Bandwidth is improved by: Faster switches: transistors are faster than wall-mounted light switches shorter wires better conductors (gold is better than copper) higher PD faster medium: fiber optics are faster than copper, radio is faster still. Additionally, one cannot see from the diameter of a cable how much copper's inside: coax has shielding that's nothing to do with signal transfers. Whatever the copper content of the cables, it's less than is in the traces on my M/b and I've no doubt the traces are faster than the cables. === Subject: RE: Acceptable number of collisions? (NOT SO!) From: "KRASZEWSKI, Marcin" <mkraszewsk@shl.com> Date: Tue, 1 Jun 1999 10:28:38 -0400 Jose M. Sanchez [mailto:opjose@ex-pressnet.com] wrote: [...] > Compared this to the same situation on 10baseT. 10BaseT > has separate send/receive pairs. In a normal 10baseT > network full duplex operation is enabled. The point to > point transfer acks no longer cause collisions, since they > occur on the other wire pairs) resulting in data streaming > at closer to the maximum spec of the lan itself (namely > 10mbit/sec)... I do not know what "a normal 10baseT network" is in the above context, but Ethernet/802.3 is basically a half-duplex network, and hubs (or concentrators) always work in the half-duplex mode. That means that they (their ports) cannot listen and talk at the same time. What a hub actually does is it reads a received packet, and without storing it or analyzing any header information, sends it to all other ports (floods the ports). As a result each device attached to a hub sees traffic (electric signal with certain specific modulation) from all other devices on the same hub (and any other hubs in the same collision domain for that matter). Collision detection part of the Ethernet protocol takes into consideration all the packets sent by devices in a single collision domain (hence the name - collision domain <g>). Switches, OTOH, usually support both duplex modes, these days most of them autosense the mode of a NIC attached to it. Each switched port, and devices attached to it, constitute a separate collision domain. Full duplex actually means that the collision detection is not implemented (or is ignored), so the NIC blindly sends packets on one pair, while listening on the other. NICs in PCs can usually support only half-duplex mode, servers usually come with NICs which can support both duplex modes, and have three settings: full, half and auto. Again, usually, not always. ==== Subject: Re: Acceptable number of collisions? From: brian davison <bdavison@proaxis.com> Date: Tue, 01 Jun 1999 23:21:07 -0700 > Bandwidth is improved by: > Faster switches: transistors are faster than wall-mounted light switches > shorter wires > better conductors (gold is better than copper) > higher PD > faster medium: fiber optics are faster than copper, radio is faster still. > > Additionally, one cannot see from the diameter of a cable how much > copper's inside: coax has shielding that's nothing to do with signal ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ actually it has lots to do with the signal.... the shielding keeps out the extraneous noise and keeps the signal in. That means you can put lots more signal down the pipe without interfering with everything else. The capacitance is what limits the frequency/distance equation and loads down the signal, if it isn't being lost as a radiated radio transmission. The switch to 10bt probably has more to do with the limiting distance, and the ability to put signal regeneration into the hubs to regain what is permanently lost in the 10b2 daisy chain. The coax could easily carry 100 mhz. or 250mhz. but for successively shorter distances, so the daisy chain configuration killed it. *** >transfers. Whatever the copper content of the cables, it's less than is in >the traces on my M/b and I've no doubt the traces are faster than the >cables. ... yip ... but look at the short distance the traces run, and the years of engineering that went into keeping the motherboard from being nothing but a major transmitting antenna. actually the motherboards are just now being used at 100 mhz. and only for the area required to tie the processor to the memory bus. everything else is at 66 mhz, 33 mhz,(pci bus for example), or 8 mhz(AT bus). brian :) ******* === Subject: Re: [OT]Electronics (Was: Acceptable number of collisions?) From: brian davison <bdavison@proaxis.com> Date: Wed, 02 Jun 1999 00:30:01 -0700 <snip p p p ed> > As far as the 10/100Mb goes, that is the data transfer > rate (maximum). I doubt that this is baseband in very > many circumstances. In fact, this is probably modulated > onto an RF carrier. By Nyquist's sampling theorem, this > carrier must have a _MINIMUM_ of 2x the data rate to avoid > aliasing of data. I would guess that the carrier > frequency is more along the lines of 4 to 5 times the > intended data rate. < snip> The 10 or 100 mbit rate is a digital signal rate, and is the data rate. Were the data encoded into an analog signal, the above would be true. as it is, the bits required for transmission are expanded by the packetizing overhead but not much else. maybe someone will figure out how to get the same kind of data rates on ethernet as the new modems do on the phone system, where a bandwidth of less than 8kbits is used to pass 56 kbits along.... were that true, then 100 mbit (100 bt) could be almost gigabit capable.... === Subject: Re: Acceptable number of collisions? (NOT SO!) From: "Richard W. Gowen" <richard@gowen.net> Date: Wed, 2 Jun 1999 16:51:59 -0500 Jose M. Sanchez <opjose@ex-pressnet.com> wrote: > Manchester encoding on a carrier frequency... > The carrier is "broadcast" to all stations over the > wire. The electrical signals exhibit the same propagation > characteristics as a typical antenna segment... > Hence the need for terminators in a 10base2 wire ... There is a big difference between an analog signal such as you see in an antenna's electromagnetic wave and a digital signal like is used in 10baseT or 10base2. There is a carrier frequency which is 10Mhz (regardless of media) but it is a digital carrier. The only resemblance an ethernet network has to a radio antenna is that the nic cards contain transceivers that send and receive signals. The signals are not the same. Don't believe me, then buy an Oscope. You'll see a square wave instead of a rounded one. And a terminator does exactly that, it terminates the square wave so it doesn't bounce back and be seen again by stations on the wire. Take one off and watch it, you'll see what I mean. > > Since you want to keep this discussion to collisions why are we talking > > about acknowledgments? > > > Because it is a factor contributing to the number of > collisions occuring in a 10base2 network. The number of collisions occuring on ANY 10base media is directly proportional to the segment utilization. It has NOTHING WHATSOEVER to do with the media type. Both 10base2 and 10baseT operate at 10Hz and have the same collision domain. The only real difference between the two are the distance limitations of the cables (100meters for 10baseT and 185meters/607ft for 10base2). This is due to the IEEE requirement of a maximum signal loss source to destination of 11.5db. These numbers are based on unrepeated segments, since a repeater will take care of signal loss. > > A 10baseT HUB is a bus topology, that is each station's > > CAT3 or better cable is physically wired to a central backing. > Not so, -ALL- 10baseT hubs have active conditioning > circuitry... again show me one that doesn't. I can do better then that. I can show you how to make your own. It's very similar to how you make a crossover cable. You take however many female RJ45 connectors as you want you hub to hold. Then connect pin1(tx+) on each connector to pin3 (rx+) on all the other connectors (one to many scenario). Then connect pin2(tx-) on each to pin6(rx-) on all the others. There you go. And it works. I built it in the lab today and tested it. Attached is a graphic if you're still confused. FYI, this is called a bus topology which still works with 10baseT even though most hubs use a repeated star topology. Just because it's seldom done doesn't mean it's against the rules. > The spec mandates this. > When I get a chance I'll find them for you... You say this but I still don't see it. That is unless your looking at the repeater requirements. But that would not apply here to an unrepeated segment. I am not just making this statement alone. Your statements made me check with others. Namely, our support engineers at both Cabletron and Cisco. Neither of them knew of this conditioning standard you speak of. Other then the db loss limitations and repeater rules, none of us have found anything on this. If you do know of some spec that applies to a single non-repeated segment and signal conditioning please give me the spec number. If not, drop it. > > > --- > > > Irrelevant, we were talking about throughput vis-a-vis > > > collisions on an ethernet lan.... > > It is relevant since we are talking about the physical layer. > > The collision domain can be set up logically in a smart hub > > but only needs to be set up physically. That was my point. > > > Eh no, your point was that there was no difference when > moving from 10base2 to 10baseT... > The actual mechanics are not in question... Yes the mechanics are in question. What detects a collision? The transceiver of a nic card. Does this transceiver handle collision detection differently based on cable type? NO. Is there a difference in collision mechanics or percentage of collisions based solely upon cable type? NO. Is there a need to mention 10base2 vs.10baseT in a discussion related to collisions? NO. One final point: Did you ever think that the reason the people you've seen migrate from 10base2 to 10baseT seem to work better is because they are moving from a bus topology to a repeated star topology? This alone makes a big difference. And this could probably change the traffic patterns on that network, thus influencing the percentage of collisions. But media type alone has no direct affect upon the number of collisions on a single segment. === Subject: Re: Acceptable number of collisions? (NOT SO!) From: "Jose M. Sanchez" <opjose@ex-pressnet.com> Date: Thu, 3 Jun 1999 00:51:41 -0400 Richard W. Gowen <richard@gowen.net> wrote: > Jose M. Sanchez <opjose@ex-pressnet.com> wrote: > > Manchester encoding on a carrier frequency... > > The carrier is "broadcast" to all stations over the > > wire. The electrical signals exhibit the same > > propagation characteristics as a typical antenna > > segment... > > Hence the need for terminators in a 10base2 wire ... > There is a big difference between an analog signal such as > you see in an antenna's electromagnetic wave and a digital > signal like is used in 10baseT or 10base2. There is a > carrier frequency which is 10Mhz (regardless of media) but > it is a digital carrier. The only resemblance an ethernet > network has to a radio antenna is that the nic cards > contain transceivers that send and receive signals. The > signals are not the same. Don't believe me, then buy an > Oscope. You'll see a square wave instead of a rounded > one. Re: Square waves: no argument from me here... > And a terminator does exactly that, it terminates the > square wave so it doesn't bounce back and be seen again by > stations on the wire. Take one off and watch it, you'll > see what I mean. > > > Since you want to keep this discussion to collisions > > > why are we talking about acknowledgments? > > Because it is a factor contributing to the number of > > collisions occuring in a 10base2 network. > The number of collisions occuring on ANY 10base media is > directly proportional to the segment utilization. It has > NOTHING WHATSOEVER to do with the media type. Theoretically no, but in actual practice the same number of workstation running on a 10base2 lan are far slower than when the same workstations are switched over to 10baseT... Again this is due to the separated RX/TX pairs... > Both 10base2 and 10baseT operate at 10Hz and have the same > collision domain. The only real difference between the > two are the distance limitations of the cables (100meters > for 10baseT and 185meters/607ft for 10base2). This is due > to the IEEE requirement of a maximum signal loss source to > destination of 11.5db. These numbers are based on > unrepeated segments, since a repeater will take care of > signal loss. > > > A 10baseT HUB is a bus topology, that is each > > > station's CAT3 or better cable is physically wired to > > > a central backing. > > Not so, -ALL- 10baseT hubs have active conditioning > > circuitry... again show me one that doesn't. > I can do better then that. I can show you how to make > your own. It's very similar to how you make a crossover > cable. You take however many female RJ45 connectors as > you want you hub to hold. Then connect pin1(tx+) on each > connector to pin3 (rx+) on all the other connectors (one > to many scenario). Then connect pin2(tx-) on each to > pin6(rx-) on all the others. There you go. And it works. > I built it in the lab today and tested it. Attached is a > graphic if you're still confused. FYI, this is called a > bus topology which still works with 10baseT even though > most hubs use a repeated star topology. Just because it's > seldom done doesn't mean it's against the rules. To paraphrase: Cool! OK, I'll concede the point... you don't need it, never the less, you probably want signal conditioning or active hubs... > > The spec mandates this. > > > When I get a chance I'll find them for you... > > > > --- > > > > Irrelevant, we were talking about throughput vis-a-vis collisions on > an > > > > ethernet lan.... > > > > > > It is relevant since we are talking about the physical layer. > > > The collision domain can be set up logically in a smart hub > > > but only needs to be set up physically. That was my point. > > > > > Eh no, your point was that there was no difference when > > moving from 10base2 to 10baseT... > > The actual mechanics are not in question... > Yes the mechanics are in question. What detects a > collision? The transceiver of a nic card. Does this > transceiver handle collision detection differently based > on cable type? NO. Is there a difference in collision > mechanics or percentage of collisions based solely upon > cable type? NO. Is there a need to mention 10base2 > vs.10baseT in a discussion related to collisions? NO. > One final point: Did you ever think that the reason the > people you've seen migrate from 10base2 to 10baseT seem to > work better is because they are moving from a bus topology > to a repeated star topology? This alone makes a big > difference. And this could probably change the traffic > patterns on that network, thus influencing the percentage > of collisions. But media type alone has no direct affect > upon the number of collisions on a single segment. Except they are NOT moving to a "repeated" star topology... Take the very same setup you created above... Transfer files ON THE SAME NICS (hopefully you have dual mode 10base2/10baseT nics...) in both modes. You'll get far better results on the 10baseT setup you're talking about above... Toss any other minor form of activity, (i.e. winblows polling the lan for workstation) and the degradation increases more than linearly... Unless by repeated star, you mean your own "passive" hub (which you insisted need not be active, hence not "repeated" either), then the results will violate the same specifications you are quoting... In effect, we should NOT see any difference between the two topologies, going by what you are saying or implying... But we do, and it's a BIG, BIG difference... === Subject: RE: Acceptable number of collisions? (NOT SO!) From: KThorpe <kthorpe@pricetrak.com> Date: Thu, 3 Jun 1999 9:59:00 +0100 Sorry, I was trying to keep out of this but I've just had to get involved. Switching our 10Base2 to 10baseT with 18 stations over three floors gave a significant increase in speed and made collisions a rarity rather than the norm. We've since had to go to switched 100MHz but that's another story. Part of the improvement is almost certainly due to improved cabling. 10base2 is very prone to damage through kinks and yanked connectors. Often this damage doesn't stop the network but does add to collisions and bad packets. I assume this is due to stray echoes on the cable from the resultant bad joints. Another part of the improvement is that AFAIK, standard 10baseT hubs are actually a star configuration and simultaneously broadcast incoming packets on all connections. This reduces the overall length of the network considerably and reduces the window during which a collision can be initiated. With ethernet, shorter is better. Long networks are much better built using token passing protocols such as Token Ring or Cambridge ring. === Subject: Re: [OT]Electronics (Was: Acceptable number of collisions?) From: Tom Burke <tomii@erols.com> Date: Thu, 03 Jun 1999 19:58:41 -0400 > At 02:11 PM 6/2/99 -0500, you wrote: > >Hmmm. All this time I thought it was 300E6 m/sec.... brian davison wrote: > It says that in my text book too!!! thats ~=~ close to > one foot per nanosecond for electron/wave propagation on > wire... so if your memory module is 6in. away from the > processor it adds at least 1 nanosecond to the access > time! ok enough trivia now It's wave propagation, not electron propagation... This is a subtle point that most people don't get (and to most people, it doesn't make a difference). Like I said, the easiest way to look at it is a garden hose full of marbles. Push one in, and immediately, one falls out the other end. The marbles are moving very slowly, but the signal travels pretty danged quickly. In reality, (depending on the media) for the voltages & currents in your computer, the electrons are moving a couple hundred miles per hour at most (I suppose I could figure it out, but it's just not worth my time right now :) )... Again, depending on the media, the wave (signal) propagates slightly slower than the speed of light. At one time, I interviewed with Intel - they wanted me on the P-II team. If I had taken the job, it would have been as an analog engineer on that little board the processor & cache sit on. They're pushing signals on it so quickly, that everything becomes transmission line theory, & impedance matching of the motherboard to the carrier to the chip is really critical. === Subject: Re: adaptec 7895 not recognized? From: Jan Carlson <janc@iname.com> Date: Thu, 17 Jun 1999 13:06:57 -0400 Don Brown wrote: > It seems that neither 5.2 or 6.0 Redhat support the > adaptec 7895 or 7896 scsi adapters. Those are definitely supported. > When I try to install Redhat on my other system, it > attempts to scan an imaginary 2940 and freezes. When I tried to install > via expert install, I found that the 7895 was not an option. This is on > a S1836DLUAN GX Thunder 100. http://www.redhat.com/hardware tells you which module to use. Look under Red Hat 6, Intel, Scsi Cards. ===