Anyone successfully bonding 2 ether ports? - Suse

This is a discussion on Anyone successfully bonding 2 ether ports? - Suse ; I have the concept successfully working using an Nvidia mobo running openSuse 11 (thanks to a great article @ Linux.com (great bridging tips for admin usage as well) plus the addition 2 software rpms I mooched from Nvidia techs which ...

+ Reply to Thread
Results 1 to 4 of 4

Thread: Anyone successfully bonding 2 ether ports?

  1. Anyone successfully bonding 2 ether ports?

    I have the concept successfully working using an Nvidia mobo running openSuse
    11 (thanks to a great article @ Linux.com (great bridging tips for admin
    usage as well) plus the addition 2 software rpms I mooched from Nvidia techs
    which were no where to be found on the site (which I can forward to you if
    you ask here).

    Should work for any motherboard/distro.

    My one trouble is I cannot maintain that config on restart and have to do the
    typing all over, post restart. (Been getting mucho config errors lately in
    openSuse 11.0...)

    If you need help tying two ports, I could probably help by steering you to
    good info and a quicky cheat I tried that works.

    and/or

    If you can help me maintain my happy config on each reboot, I'd welcome
    it. (Been getting mucho config errors in openSuse 11.0...)

    Is it faster to have 2 ports? Not sure, but if your I/O is being sliced an
    diced over two ports, you have to think it IS...

    Anyway, a fun project for someone with medium to light skills...

  2. Re: Anyone successfully bonding 2 ether ports?

    Hi Pete

    Re being faster..

    I am a bit rusty on this, but if you setup the box with 2 ports
    configured with (say) one IP address your connected layer 2 switch also
    needs some intelligence to know that you are running it this way. Cisco
    L2 switches for example allow you to combine their physical ports for
    backbone and higher bandwidth situations.

    The basic problem is ARP or the broadcast/learn facility whereby a
    hardware (usually MAC) address is married up with the IP address. A more
    simple L2 switch will get confused that the IP address is appearing on
    two of its ports and might even just continually switch its own ARP
    table between them! From memory Cisco's CDP use to get into trouble and
    had to be disbled if you setup a "loop" between two of their devices (eg
    connecting two L2 switches with two cables) in this way.

    And true to form I cant dredge up the name/function that was used to
    describe/configure this. It's been a while! I use to run a 100 user
    enterprise that had 5 Cisco switches backboned with two cables per
    master<>remote connection and two connections from each main server to
    the master switch. Never did do any measurements to check whether it was
    faster, it just seemed an intelligent thing to do at the time because
    the LAN had a lot more network I/O than the "usual" office environment.

    Another thing to keep in mind is that ethernet cards in PC's may require
    a fair slice of CPU grunt to do things. I remember that the first
    "older" 1GB cards could only really sustain 200-300MB/sec for this
    reason. The fix at the time was to buy ethernet hardware that had its
    own onboard processor to push the data into memory in DMA mode, thus
    reducing the main CPU overhead. Whether this is still the case nowadays
    I dont know, Ethernet specs never seem to publish throughout, only sync
    speed!

    Just some things to keep in mind..

    Cheers Bob

    Pete Puma wrote:
    > Is it faster to have 2 ports? Not sure, but if your I/O is being sliced an
    > diced over two ports, you have to think it IS...
    >
    > Anyway, a fun project for someone with medium to light skills...


  3. Re: Anyone successfully bonding 2 ether ports?

    Thanks for writing--I enjoyed reading this...

    Bob Bob wrote:

    > Hi Pete
    >
    > Re being faster..
    >
    > I am a bit rusty on this, but if you setup the box with 2 ports
    > configured with (say) one IP address your connected layer 2 switch also
    > needs some intelligence to know that you are running it this way. Cisco
    > L2 switches for example allow you to combine their physical ports for
    > backbone and higher bandwidth situations.


    Yikes--I'm just banging signals to my simple home Linksys router, with a few
    gigabit routers off to the sides for video transfers (which I cannot get
    quicker than 12 megs per sec transfer--but that's for another day).
    I doubt my router has the logic to sense this "experiment", although I have
    seen many
    different pulse "beats" between the two eth lights--makes it *look* like
    they know what they're doing (at the risk of getting beat up over that
    assumption). I could use a test or tool of some kind after bonding to see
    what the T'fer is....

    > The basic problem is ARP or the broadcast/learn facility whereby a
    > hardware (usually MAC) address is married up with the IP address. A more
    > simple L2 switch will get confused that the IP address is appearing on
    > two of its ports and might even just continually switch its own ARP
    > table between them! From memory Cisco's CDP use to get into trouble and
    > had to be disbled if you setup a "loop" between two of their devices (eg
    > connecting two L2 switches with two cables) in this way.


    In openSUSE 11, you can disconnect the ports specifically for the use of
    bonding (there's even a button for it). Assuming you have Yast install the
    "ib-bonding" - infiniband bonding tool
    loaded and turn off Network Manager on your ports for the old school ifup.

    Post setup, ifconfig shows only one MAC for your newly made "bond0" while the
    specs for eth0 & eth1 have no more MAC address nor any IP address assigned
    (which is what you tell it to do in Yast). At this point I am still stuck
    reconfiguring it every boot--if I have the ETHs come on in boot, they get
    assigned an address and you must then ifdown both of them, enslave, then run
    the bond0.

    At that point, I don't know what a 60-buck router is thinking (flashed with
    DD-WRT software, which is much more versatile, but smarter--I don't know.

    > And true to form I cant dredge up the name/function that was used to
    > describe/configure this. It's been a while! I use to run a 100 user
    > enterprise that had 5 Cisco switches backboned with two cables per
    > master<>remote connection and two connections from each main server to
    > the master switch. Never did do any measurements to check whether it was
    > faster, it just seemed an intelligent thing to do at the time because
    > the LAN had a lot more network I/O than the "usual" office environment.
    >
    > Another thing to keep in mind is that ethernet cards in PC's may require
    > a fair slice of CPU grunt to do things. I remember that the first
    > "older" 1GB cards could only really sustain 200-300MB/sec for this
    > reason. The fix at the time was to buy ethernet hardware that had its
    > own onboard processor to push the data into memory in DMA mode, thus
    > reducing the main CPU overhead. Whether this is still the case nowadays
    > I dont know, Ethernet specs never seem to publish throughout, only sync
    > speed!


    I have seen some positively EVIL looking PCI ether cards advertised in CPU
    magazine--probably it's my first look at commercial boards that has me so
    fascinated--but you could communicate with the space shuttle with these
    things in real time. (Course, that would be a long Cat 6 wire...)

    > Just some things to keep in mind..
    >
    > Cheers Bob


    Thanks for posting. I'm having fun playing with this. Using an overclocked
    dual proc Intel, so I feel no pains in the processing department right now.

    The cool source for this that got me started was
    http://www.linux.com/feature/133849
    ---there's info that admins would find really informative. If you bridge
    (bridge0) your ports as opposed to bonding, you can customize your isolation
    from port to port, etc. If you're serving a small net, you definitely
    want to know some of these tricks. None of the magazines have talked about
    this in well over 2 years that I've seen, which is about how long consumer
    dual port motherboards have been around (server boards probably longer).

    and there's more, not surprisingly.
    Ultimately, I'm not sure if there will be any payoff in doing this,, but it's
    what Linux is about, for me...





  4. Re: Anyone successfully bonding 2 ether ports?

    Hi Pete

    I wanted to take more time to answer this but I am getting kind of busy...

    Testing the system? I havent actually looked at the process you went
    through. Some linux methods though create two "physical" interfaces (eg
    eth0 and eth1) and then a combined one. You can then look at the iface
    stats for each, or there may be some tools that came with the binding
    code...

    Failing that a flood ping (ping -f -s 1500 hostname - as root) to some
    host that has a faster or same connection type on your switch, or if it
    is (single hop to) a router in its own right the use that instead. That
    will max out the interface bandwidth so if you experiment by removing
    cables one at a time you can see the results. You can also do this to
    check CPU usage if you suspect a card is using a lot.

    LEDs? Not really reliable. You dont know if the flashes are real data or
    repeated blocking! You may however detect a pattern in the flood ping
    experiment...

    Back to work!

    Cheers Bob

    Pete Puma wrote:
    > Yikes--I'm just banging signals to my simple home Linksys router,

    with a few
    > gigabit routers off to the sides for video transfers (which I cannot get
    > quicker than 12 megs per sec transfer--but that's for another day).
    > I doubt my router has the logic to sense this "experiment", although I have
    > seen many
    > different pulse "beats" between the two eth lights--makes it *look* like
    > they know what they're doing (at the risk of getting beat up over that
    > assumption). I could use a test or tool of some kind after bonding to see
    > what the T'fer is....


+ Reply to Thread