aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) (09/29/89)
Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, I am listing what I know about the topic: THINWIRE Flexible Low cost ( app. $3.00 per meter ) 10 Mb bandwidth Max segment length - 185 meters (30 nodes per segment) One multiport repeater can handle upto 8 segments THICKWIRE More resilient for running through floors and ceilings Higher cost (app $11.00 per meter) 10 Mb bandwidth Max segment length - 500 meters Based on the above, I would choose thickwire ONLY if the length of the segment had to be more than 500 mts or if the wire was going to run through adverse areas. Any comments ?? -vikas ------------------------------------------------------------------------- JOHN VON NEUMANN NATIONAL SUPERCOMPUTER CENTER INTERNET: aggarwal@jvnca.csc.org UUCP: rutgers!jvnca!aggarwal BITNET: aggarwal@jvncc --------------------------------------------------------------------------
jacob@gore.com (Jacob Gore) (10/01/89)
/ comp.protocols.tcp-ip / aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) /Sep 29/ [lists some tradeoffs between thin and thick wire Ethernet] Also: THINWIRE Transceivers and drop cables can be used Transceiver can be (and often is) built into the computer instead If using transceivers and drop cables: Non-intrusive transceivers are not available (cable must be cut to insert new transceiver) If not using transceivers and drop cables: Cable generally runs within an inch of each computer Non-intrusive off-shoot cables are available, but their outlets must be inserted into the cable in advance, and they add twice their length to the total length of the network THICKWIRE Transceivers and drop cables must be used Non-intrusive transceivers are available ("vampire-tap") In short, if your network is best served by a central-spine-with-drop-cables topology, then one additional advantage of thick Ethernet is that you can add new transceivers easily. And then there's Twisted-Pair Ethernet... Jacob -- Jacob Gore Jacob@Gore.Com boulder!gore!jacob
kratz@bnrgate.bnr.ca (Geoff Kratz) (10/02/89)
In article <8909291306.AA06775@jvnca.csc.org>, aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > > Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, What about Unshielded Twisted Pair (UTP)? We are using this exclusively now, and we have found a number of advantages: - extremely cheap (uses existing 4-wire in the building) - MAUs cost much less than transceivers - You can use the BIXX cross-connects to simplify moves of workstations (a move requires moving jumper wires from one cross-connect to another) The disadvantage, of course, is the distance. UTP is only good to about 400 meters. Within a building, though, this is usually adequate. Another thing is the tolerance on the clock crystals. We found a number of workstation manufacturers who's specs on the clock were less than adequate, thus causing a lot of jams or CRC's. However, for our site (1000+ workstations in 5 buildings with constant workstation moves), UTP gives us more benefits in topology planning, installation and $$$. And yes, we run at full ethernet (10 Mbps). Anyone else out there using UTP for their ethernets? -- Geoff Kratz Bell-Northern Research, Ltd. Ph: (613) 763-5784 Internet Systems P.O. Box 3511, Station C FAX:(613) 763-3283 Ottawa Ontario Canada K1Y 4H7 BITNET: kratz@bnr.ca
eric@lcc.la.Locus.COM (Eric Peterson) (10/02/89)
In article <8909291306.AA06775@jvnca.csc.org> aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > >Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, >I am listing what I know about the topic: > >THINWIRE > > Flexible, Low cost ( app. $3.00 per meter ), 10 Mb bandwidth, > Max segment length - 185 meters (30 nodes per segment), One > multiport repeater can handle upto 8 segments From past experience, I recommend this for only very small or temporary nets. The cable is fragile, and the cable -> BNC connection is especially susceptible to breakage (we once had a janitor vacuuming the floor accidentally hit the cable and yank it out of the BNC connector, he noticed the damage and managed to push the cable back into the connector - needless to say, it was a poor connection and that whole floor of the building suffered from intermittent Ethernet problems for several days while we looked for the problem). >THICKWIRE > > More resilient for running through floors and ceilings, Higher > cost (app $11.00 per meter), 10 Mb bandwidth, Max segment > length - 500 meters > >Based on the above, I would choose thickwire ONLY if the length of the >segment had to be more than 500 mts or if the wire was going to run >through adverse areas. I feel that any area with users in it is "adverse." >Any comments ?? >-vikas Another solution is to use multiport transceivers to disribute the Ethernet. It is much less susceptible to damage (or I should say, if damage _does_ occur, it doesn't propagate throughout the network.) I believe the spec allows xcvr cable runs up to 50 meters away, and the multiports that I've seen allow up to two levels of stacking the xcvrs. Another advantage is that some Ethernet boards do not have the built-in thin wire BNC connector. A disadvantage is that manufacturers haven't seem to standardize on their method of attaching the 15 pin connector (mechanically) to the board (grrrr). Eric -- Eric Peterson, Locus Computing Corporation lcc.eric@seas.ucla.edu; {randvax,sdcrdcf,ucbvax,trwspp}!ucla-se!lcc!eric {gryphon,turnkey,attunix,oblio}!lcc!eric (213) 337-5153
kincl@IAG.HP.COM (Norm Kincl) (10/04/89)
You forgot to mention twisted-pair (IEEE 802.3 10BaseT). In some situations---especially office type environments---this can be the best solution. I forget the exact specs on it, but we run twisted pair from a patch pannel to all the offices. In addition to not having to have to string coax around to everyone (just pull it along with the phone wire), it gives us the flexibility to quickly patch any office into any subnet or bridged segment. This is useful for us as people constantly move their office, yet not the disckless cluster they are a part of. -Norm Kincl Informatin Architecture Group Hewlett-Packard
melanie@ph-meter.beckman.uiuc.edu (Melanie Anderson) (10/05/89)
In article <68@lcc.la.Locus.COM>, eric@lcc.la.Locus.COM (Eric Peterson) writes: > > In article <8909291306.AA06775@jvnca.csc.org> aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: enough thinwire bashing, already. im coming into this debate late; but let me throw in my 2pence anyhow: some background: the beckman institute is a research lab/think tank at the uofi. we have over 300,000 sqare feet on five floors plus a basement. the building is designed to house over 700 researchers in all types of research settings from offices to computer labs to chemistry and mri. the building is not yet at capacity and we already have over 300 computers on site ranging from zenith laptops to vaxes, ardents and apollos to a 32K processor connection machine. the design goal of the network was to provide high connectivity to every office, lab, conference room and public area. therefore every room in the building (excluding bathrooms and kitchens!) has at least one connection to the building network. this presented some interesting cable design problems, resulting in (if you dont belive me do the math) the discovery that it would be 1. enormously expensive 2. very difficult 3. would not meet connectivity requirements to use thick ethernet. the entire building is wired with thinwire ethernet, close to 16 miles of it, in fact. the cabling design is a rib-and-spine approach. areas of the building were carved into logical ethernets (ribs) and each rib connects, via cabletron thin repeaters, to a 100Mbit beckbone (good pun, huh?) built out of network systems en641 routers. it WORKS, it works GREAT. i have had zero problems with the wiring, ~zero problems with the repeaters (out of over 700 repeater ports i have had 1 fail). somebody comes in, they get a chunk of cable from central stores for 6 bucks, they come to me, i give them a new IP number and tell them what their default gateway ip address is (for those brain-dead ips out there!), they go up to their office, they plug in, and off they go! in fact, this wiring scheme and these repeaters have worked so well the same system is being used in a 400-port new installation in the psychology building on campus, in the cedar project addition (100+ ports) and will probably be used in other new construction. > > > >Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, > >I am listing what I know about the topic: > > > >THINWIRE > > > > Flexible, Low cost ( app. $3.00 per meter ), 10 Mb bandwidth, > > Max segment length - 185 meters (30 nodes per segment), One > > multiport repeater can handle upto 8 segments where are you buying your repeaters from? here at beckman i have 33 thinwire repeaters purchased from cabletron. they support anywhere from 12 thin and two thick connections to 84 thin segments and two thick. also, if you are using a repeater, you only get 28 nodes a segment. the plus to using thin repeaters over conventional thick is that most reapeaters are smart enough to shut down a port if it detects a bizzare electrical condition like a short, open or fluctuation in resistance. in exremely large networks, this intelligence can save hours or days of time. it also fixes the problem below. > > From past experience, I recommend this for only very small or > temporary nets. The cable is fragile, and the cable -> BNC > connection is especially susceptible to breakage (we once had > a janitor vacuuming the floor accidentally hit the cable and > yank it out of the BNC connector, he noticed the damage and > managed to push the cable back into the connector - needless > to say, it was a poor connection and that whole floor of the > building suffered from intermittent Ethernet problems for > several days while we looked for the problem). > first, i would bet that a thick transciever cable would be damaged by this kind of rough treatment, as well. also, you might want to suggest that if someone breaks something, they report it, rather than try to fix it... second, i would suggest you investigate who is doing your bnc terminations. i have over 2000 terminations in this building and very few (~25) have been demolished, usually by furniture being set on them. i would argue that you would have the same problem with el-cheapo thick connections. i do have wall and floor boxes installed. NO wiring should lie on a floor, be it rs-232, power, ether or piano. > >THICKWIRE > > > > More resilient for running through floors and ceilings, Higher > > cost (app $11.00 per meter), 10 Mb bandwidth, Max segment > > length - 500 meters > > > >Based on the above, I would choose thickwire ONLY if the length of the > >segment had to be more than 500 mts or if the wire was going to run > >through adverse areas. > thin ethernet with a toughened-up pvc or teflon coating is available for adverse-condition or plenum applications. like i said, even our long underfloor and overhead runs are thinwire. thick is appropriate for certain applications. i strongly believe, however, that whatever your wiring application, that ribs-n-spine be used in any new large network. saves you much time and effort and grief. it is flexible, reconfigurable, and expandable. try to divvy your network up into ribs and then firewall them off from each other so that one looney machine wont crash everyone. whatever you do, PLAN for cut wires, smushed connectors, gonzo transcievers, crazy ethernet cards, and broken software. DONT assume all computers will play well with others. expect the worst because it WILL happen. > > Another solution is to use multiport transceivers to disribute > the Ethernet. It is much less susceptible to damage (or I should > say, if damage _does_ occur, it doesn't propagate throughout the > network.) I believe the spec allows xcvr cable runs up to 50 > meters away, and the multiports that I've seen allow up to two > levels of stacking the xcvrs. Another advantage is that some > Ethernet boards do not have the built-in thin wire BNC connector. > yeah, you have to buy a transciever for any machine that doesn have an on-board transciever. the thin transcievers are the same price, or cheaper, than thick. you can also put a delni-alike on thinwire ethernet (see above.) Melanie Anderson melanie@ph-meter.beckman.uiuc.edu Beckman Institute 217/244-1079 Unversity of Illinois ----------------------------------------------------------------------------- "N4634X, Champaign Tower, uhh, what's that hanging off your wing?" -----------------------------------------------------------------------------
kratz@bnrgate.bnr.ca (Geoff Kratz) (10/05/89)
Before I start, I want to say this I do not want to bash thin-net. The purpose of this is more to provide an alternative for those people who are now (will soon be) building a LARGE ethernet site. In article <1989Oct4.174149.9451@ux1.cso.uiuc.edu>, melanie@ph-meter.beckman.uiuc.edu (Melanie Anderson) writes: > In article <68@lcc.la.Locus.COM>, eric@lcc.la.Locus.COM (Eric Peterson) writes: > > > > In article <8909291306.AA06775@jvnca.csc.org> aggarwal@JVNCA.CSC.ORG > (Vikas Aggarwal none) writes: > > enough thinwire bashing, already. > > im coming into this debate late; but let me throw in my 2pence anyhow: > > some background: > > the design goal of the network was to provide high connectivity to > every office, lab, conference room and public area. therefore every room in > the building (excluding bathrooms and kitchens!) has at least one > connection to the building network. this presented some interesting cable > design problems, resulting in (if you dont belive me do the math) the > discovery that it would be 1. enormously expensive 2. very difficult 3. > would not meet connectivity requirements to use thick ethernet. It would have been as trivial with UTP. We are located in a number of buildings, but a prototypical one would be: 1000-2000 people about 1000+ machines (workstations, mainframes, PC's, Macs) about 4000+ ports of repeaters (on the hubs, bridges and delnis) And we move equipment around a LOT. Using UTP allowed us to put the backbone in a central location, and then build a star from there. Since we already have 4-pair cables running to the floor, wiring was cheap. We did find, though, that the cost per seat was about $800 when installation of the cable was included (that includes the prices of the twisted pair, MAU, hubs, bridges and delnis). Cheaper, of course, when the cable was already in place. And it can get as good as $400-500 per seat (depending on numbers). > it WORKS, it works GREAT. i have had zero problems with the wiring, ~zero > problems with the repeaters (out of over 700 repeater ports i have had 1 > fail). somebody comes in, they get a chunk of cable from central stores for > 6 bucks, they come to me, i give them a new IP number and tell them what > their default gateway ip address is (for those brain-dead ips out there!), > they go up to their office, they plug in, and off they go! We tried thin in a few locations, and got burned by poor construction (this is probably not true of all manufacturers thing products). Just shifting a workstation could cause the T-connector to come off the BNC, thus causing a break and taking out the entire segment. We find that with UTP, a MAU can die and have no adverse affect on the net (depends on the type of death of course. If it just fails and stops functioning no problem. If it decides to transmit junk, that's a different story). On the rare occasion that a hub decides to lose its mind, the effect can vary (depends on where the hub is located in the topology). At most, though, you will lose a segment behind a bridge. We have yet to have a bad bridge or hub take out the entire subnet within a building. > i strongly believe, however, that whatever your wiring application, > that ribs-n-spine be used in any new large network. saves you much time and > effort and grief. it is flexible, reconfigurable, and expandable. try to > divvy your network up into ribs and then firewall them off from each other > so that one looney machine wont crash everyone. Yup, same with UTP. You use hubs to build a star, use the bridges as traffic filters and use routers to build multiple backbones. As well, moves involve moving a pair of jumper wires on a BIXX block. > whatever you do, PLAN for cut wires, smushed connectors, gonzo > transcievers, crazy ethernet cards, and broken software. DONT assume > all computers will play well with others. expect the worst because it > WILL happen. You betcha! And expect some of the users to get "creative" too. Large sites do present an interesting problem and an incredible opportunity to do it right (or screw it up REAL GOOD!) whether it is thick, thin or UTP. -- Geoff Kratz Bell-Northern Research, Ltd. Ph: (613) 763-5784 Internet Systems P.O. Box 3511, Station C FAX:(613) 763-3283 Ottawa Ontario Canada K1Y 4H7 BITNET: kratz@bnr.ca
glen@aecom.yu.edu (Glen M. Marianko) (10/08/89)
In article <8909291306.AA06775@jvnca.csc.org>, aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > > Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, > I am listing what I know about the topic: > > THINWIRE > ... > Max segment length - 185 meters (30 nodes per segment) This posting reminded me of a question I had about thin ethernet and the 30 nodes per segment limit. Why is this? I thought on thin ethernet you can have a node every .5 meters (vs. thick which is marked for much more). If so, then 185/.5=370 according to my calculator with the weak battery :-). So why couldn't I have more than 30 workstations? Tell ya, the only place I ever saw this in print was in a DEC catalog. Maybe this is a DEC restriction? -- -- Glen M. Marianko Manager, LAN Services Glasgal Communications, Inc. 151 Veterans Drive Northvale, New Jersey 07647 201-768-8082 glen@aecom.yu.edu - {uunet}!aecom!glen (Courtesy of AECOM & unaffiliated)
ian@lassen.wpd.sgi.com (Ian Clements) (10/09/89)
In article <2525@aecom.yu.edu>, glen@aecom.yu.edu (Glen M. Marianko) writes: > In article <8909291306.AA06775@jvnca.csc.org>, aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > > > > Max segment length - 185 meters (30 nodes per segment) > > ...than 30 workstations? Tell ya, the only place I ever saw this in > print was in a DEC catalog. Maybe this is a DEC restriction? Sorry, see section 10.7.1 of the IEEE Std. 802.3-1988, page 39. It says "A coax segment may contain a maximum of 185 m (600 ft.) of coaxial cable and a maximum of 30 MAUs." Cheers, Ian
pat@hprnd.HP.COM (Pat Thaler) (10/14/89)
> / hprnd:comp.protocols.tcp-ip / kratz@bnrgate.bnr.ca (Geoff Kratz) / 11:11 am Oct 1, 1989 / > In article <8909291306.AA06775@jvnca.csc.org>, aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > > > > Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, > > What about Unshielded Twisted Pair (UTP)? We are using this exclusively > now, and we have found a number of advantages: > > - extremely cheap (uses existing 4-wire in the building) > - MAUs cost much less than transceivers > - You can use the BIXX cross-connects to simplify moves of workstations > (a move requires moving jumper wires from one cross-connect to another) > > The disadvantage, of course, is the distance. UTP is only good to about > 400 meters. Within a building, though, this is usually adequate. Another The design goal for 10BASE-T is 100 m on 24 AWG UTP. 400 meters is excessive. > thing is the tolerance on the clock crystals. We found a number of > workstation manufacturers who's specs on the clock were less than adequate, > thus causing a lot of jams or CRC's. However, for our site (1000+ The clock problem has to do with the effect of excessive clock skew on repeaters, not with any characteristic of twisted pair vs coax (except that with coax you may not have had to have a repeater between your nodes). Repeaters decode the incoming signal and reencode it with their internal clock in order to remove jitter. The clocks of the sending DTE and the repeater may be slightly different, but the must be within 10 MHz +- 0.01% to meet the spec. A device with a fast clock can transmit a maximum length packet in about 300 ns faster than a device with a slow clock. To allow for this skew, a repeater stores at least 3 bits in its FIFO before it starts transmitting (so it won't run out of bits if it is fast) and has space for at least 3 more bits to accumulate (so it won't drop bits if it is slow). If clock frequency is out of spec then packets can get corrupted. Since people switch from a small unrepeatered coax network to a twisted pair network, they discover the problem when the repeater is inserted. > workstations in 5 buildings with constant workstation moves), UTP gives us > more benefits in topology planning, installation and $$$. > > And yes, we run at full ethernet (10 Mbps). > > Anyone else out there using UTP for their ethernets? > -- > Geoff Kratz Bell-Northern Research, Ltd. Ph: (613) 763-5784 > Internet Systems P.O. Box 3511, Station C FAX:(613) 763-3283 > Ottawa Ontario Canada K1Y 4H7 > BITNET: kratz@bnr.ca > ----------
pat@hprnd.HP.COM (Pat Thaler) (10/14/89)
> / hprnd:comp.protocols.tcp-ip / glen@aecom.yu.edu (Glen M. Marianko) / 6:01 pm Oct 7, 1989 / > In article <8909291306.AA06775@jvnca.csc.org>, aggarwal@JVNCA.CSC.ORG (Vikas Aggarwal none) writes: > > > > Just to collect one's views on Thinwire ethernet vs Thickwire ethernet, > > I am listing what I know about the topic: > > > > THINWIRE > > ... > > Max segment length - 185 meters (30 nodes per segment) > > This posting reminded me of a question I had about thin ethernet and > the 30 nodes per segment limit. Why is this? I thought on thin > ethernet you can have a node every .5 meters (vs. thick which is > marked for much more). If so, then 185/.5=370 according to my > calculator with the weak battery :-). So why couldn't I have more > than 30 workstations? Tell ya, the only place I ever saw this in > print was in a DEC catalog. Maybe this is a DEC restriction? > -- It comes from the IEEE 802.3 10BASE2 standard. It is actually 30 MAUs per segment that is important, not how many stations are attached to those MAUs. It is due to the effect of the MAU (leakage current, loading, etc.) and the effect of the connectors used to attach the MAU on the media. The spacing requirement is to ensure you don't have a bunch of MAUs clustered right together as that can lead to a larger reflection. MAUs can be closer together on thin than on thick for two reasons: 1) the attenuation of thin is higher (part of the reason it only goes 185 m); and 2) the number of MAUs is smaller on thin 30 vs. 100. > -- Glen M. Marianko Manager, LAN Services Glasgal Communications, Inc. > 151 Veterans Drive Northvale, New Jersey 07647 201-768-8082 > glen@aecom.yu.edu - {uunet}!aecom!glen (Courtesy of AECOM & unaffiliated) > ---------- Pat Thaler