Switchless TCP/IP networking?

Discussion in 'Linux Networking' started by Matthew Clark, Mar 10, 2006.

  1. What if I used thinnet as the connection medium? Could I?

    1) Adapt the blades' TP ports to a BNC connector?
    2) String them together via thinnet?

    Is this feasable? Also, I've read that "100Base2" doesn't
    exist. Are they simply referring to a lack of hardware NICs,
    or a limitation of the coaxial medium itself?

    Matt
     
    Matthew Clark, Mar 21, 2006
    #21
    1. Advertisements

  2. Matthew Clark

    Jim Hicks Guest

    coaxial wire has two conductors, (1 pair), 100mbit ethernet adapters
    have four wires, two twisted pairs of wires.
    Coax has an impedence of 75ohms, or 92ohms. Twisted pair has a different
    impedence, but I forget what for now.
    Twisted pair requires the transmit pair from one adapter to connect to
    the receive pair on the other adapter, and vice versa. This allows an
    adapter to transmit and receive simultaneously.
    Coax has a tranceiver that is either transmitting, or receiving. This
    limits the speed of the conversation, and is referred to as half duplex.
    Aside from the data losses due to mismatched impedence, you would have
    to invent a tranceiver for this to work. In addition it would require a
    firmware patch on the card to implement the collision avoidance and
    collision detection that is not required on switched links.
    You are doing this to avoid buying a $30 switch????
     
    Jim Hicks, Mar 21, 2006
    #22
    1. Advertisements

  3. You are doing this to avoid buying a $30 switch????

    No I'm doing this in an environment that would fry a $30 switch...

    Matt
     
    Matthew Clark, Mar 21, 2006
    #23
  4. You are doing this to avoid buying a $30 switch????
    Oops, forgot to thank you for the info. Sorry 'bout that. :)

    Matt
     
    Matthew Clark, Mar 21, 2006
    #24
  5. Matthew Clark

    Jim Hicks Guest

    I did middleware testing in a large environment, using commercially
    available switches, more than $30, we did performance testing using an
    LDAP database with more than 20 million names and addresses, millions of
    sessions established, and never once did we have an issue that turned
    out to be the switch. We were using major brands, all major
    manufacturers of managed switches. (To verify performance with a variety
    of hardware.) If you suspect the switch, use another. Testing in a
    topology that does not include switches is not valid, unless your target
    network will not have switches, because all internet connections on the
    planet include switches in the mix.
     
    Jim Hicks, Mar 21, 2006
    #25
  6. Testing in a topology that does not include switches is not valid,
    The target network will not have a switch. Not my choice,
    wish it weren't the case, but it is what it is and I gotta make it
    work.

    Matt
     
    Matthew Clark, Mar 21, 2006
    #26
  7. Matthew Clark

    Jim Hicks Guest

    Look up some topology for FDDI. IBM had a dual card copper or fiber FDDI
    topology that daisy chained the transmit and receive through the two
    adapters giving some redundancy in case one box failed. I will try to
    dredge up my pictures of this topology. An issue of daisy chaining is
    the failure of one system/adapter taking out the whole network.
    If your target network has no switch, how is it connected? Could you use
    that topology, or are you inventing the switchless target network in
    addition to testing it?
     
    Jim Hicks, Mar 21, 2006
    #27
  8. Matthew Clark

    Jim Hicks Guest

    I lied. FDDI is a token passing protocol, more condusive to daisy chain
    topology than ethernet.
     
    Jim Hicks, Mar 21, 2006
    #28
  9. Matthew Clark

    X Guest

    Matthew,

    I don't think you got a real clear reply on the assigning of an IP
    address to bridge NICs, but here it is.

    You CAN assign an IP address to the bridge device (e.g. br0)

    You CAN NOT assign an IP address to the individual NICs (e.g. eth0 and
    eth1) used in a bridging device.

    So, in effect, you may still access the machine with an IP address if
    you like, but the IP address assignment has to be made to the bridge
    device (br0) rather than the NICs (eth0 or eth1). Also, enable STP
    (Spanning Tree Protocol) on each machine, as mentioned in a previous
    post, as this should avoid loops and account for node failures.

    If you went the routing direction, you would also have to use zebra
    (OSPF) or some other method to detect redundant routes so as to avoid
    routing loops. To my thinking, bridging seems like a more desirable
    method for this scenario.

    X
     
    X, Mar 21, 2006
    #29
  10. Matthew Clark

    Jim Hicks Guest

    Matthew Clark wrote
    have four wires, two twisted pairs of wires
    Coax has an impedence of 75ohms, or 92ohms. Twisted pair has
    different
    impedence, but I forget what for now
    Twisted pair requires the transmit pair from one adapter to connect t

    the receive pair on the other adapter, and vice versa. This allows an
    adapter to transmit and receive simultaneously
    Coax has a tranceiver that is either transmitting, or receiving. This
    limits the speed of the conversation, and is referred to as hal
    duplex
    Aside from the data losses due to mismatched impedence, you would hav

    to invent a tranceiver for this to work. In addition it would requir
    a
    firmware patch on the card to implement the collision avoidance and
    collision detection that is not required on switched links
    You are doing this to avoid buying a $30 switch???
     
    Jim Hicks, Mar 21, 2006
    #30
  11. Matthew Clark

    Cat Guest

    Indeed, it's nearly impossible to simply connect a TP card to a ThinNet
    network. The only way that I know of is by using a semi-smart device,
    such as a hub, that has both TP and BNC transceivers. Even then, things
    are ugly. I use this at home to connect my laptop (TP) to 10Base2 and I
    have really strange packet loss when talking to only certain nodes on
    the 10Base2.

    It seems my patience too has run out, so to speak.. What in the world
    are you trying to do?!

    -Cat
     
    Cat, Mar 22, 2006
    #31
  12. Twisted pair wires are commonly available/used with the following
    impedances:
    - 100 ohms (nodern sructured cabling, CAT5/CAT6 cables)
    - 110 ohms (used on some automation applications and digital audio)
    - 120 ohms (seen on telecom infrastruture)
    - 150 ohms (used on old IBM data cabling)

    Nowadays the trand is to use 100 ohms cable.
    The normal 100mbit ethernet adapters are designed to work with
    100 ohms twisted pair wiring.
     
    Tomi Holger Engdahl, Mar 22, 2006
    #32
  13. Not easily. The TP port loses information that is needed to produce the
    BNC output. Perhaps you could find cards with an AUI or MUI port on them.
    It simply doesn't exist. There is no specification, there are no NICs,
    and there is no specified coaxial medium. There are certainly coaxial media
    in existence over which one could pass data at 100Mbps, but nobody has ever
    standardized a way to do that for Ethernet.

    DS
     
    David Schwartz, Mar 22, 2006
    #33
  14. Matthew Clark

    Moe Trin Guest

    [/QUOTE]

    You'd need 10Base2 NICs

    There is no standard. The limitation is where the spacing between the
    center and outer conductors divided by the dielectric constant of the
    insulation between the conductors becomes a significant fraction (say 1/4)
    of a wavelength of the carrier. There is also a distance limitation due to
    loss increasing with carrier frequency. You _could_ stuff bits down the
    wire at 18 Gigabit, but the hardware doesn't exist, the wire would be small
    and fragile, and the maximum length is only going to be a couple of meters
    or so - call it 18000Base.003 ;-)
    except that some used four pairs
    Of the four hundred odd coaxial cables listed in MIL-C-17, over 85 percent
    are 50 Ohms. 75 Ohm (variously 70 or 72 Ohm) is much less common, but is
    the standard impedance used for cable TV and some amateur radio stuff.
    There are a very limited number of cables at impedances other than 50 or
    75 Ohm - 93 Ohm normally is used for pulse transmissions - RG-62/U or
    RG-71/U being two examples. Ethernet used a 50 Ohm cable that differed
    from RG-8/U (thicknet) or RG-58/U (thinnet) in that the jacket material
    does not create toxic gases and lots of smoke when burnt.
    Varies as a function of wire size, spacing, and the dielectric constant
    of the insulating medium, and is influenced by the proximity to other
    conductors or ground.

    Old guy
     
    Moe Trin, Mar 23, 2006
    #34
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.