My testing of how load balancing works with more than two interfaces



  • I believe (could be wrong) that I have read somewhere in the forums that load balancing only supports two interfaces.
    That is actually incorrect from the results of the testing I have done.
    I only have three internet connections a person with four may want to do some further testing to confirm or refute my findings.

    There are three main points that I have found:

    1. Opens up or initiates a new connection using round robin selecting from all interfaces.
    2. The Upstream speed is equal to the speed of the fastest of the last interface up speed and the next to last interface up speed. As added to the load balance Pool.
    3. The Downstream speed is equal to the sum of first interface down speed and the second interface down speed. As added to the load balance Pool.

    In my case this behaviour means I get the best results when I only use two interfaces, however it doesn't mean that true for everybody it is case dependant.

    Taking my current Internet interfaces as an example

    1. Cable 10Mbps Down 1.3Mbps up
    2. DSL1 10Mbps Down 550Kbps up
    3. DSL2 3Mbps Down 128kbps up

    (The figures given above are not the SLA max but the more realistic achieveable measured figures).

    using the rules I state above If I use a load balance of just 1, 2
    then I would get
    Down speed of 20Mbps Up speed of 1.3Mbps

    If I try combinations of using three interfaces I will never get any better see examples below:

    Pool Order  |  Down  |   UP   
    1,2,3         | 20M     |  550K
    3,2,1         | 13M     |  550K
    2,1,3         | 20M     |  1.3M

    Hopefully this testing can be confirmed by others and perhaps will assist people in understanding if it is worth them having more than two interfaces.
    Bare in mind that in my tests I am focussed on maximum up and down throughput I have not considered other varibles such as CAP limits or the behaviour when the CAP limit is reached i.e. additional cost or throttling to lower speeds.

    Thanks

    Paul.



  • @idkpmiller:

    I believe (could be wrong) that I have read somewhere in the forums that load balancing only supports two interfaces.
    That is actually incorrect from the results of the testing I have done.

    Links?

    Traffic shaping is currently only supported on two interfaces.
    Traffic shaping is not the same as load balancing.
    –> It can be used in each and every firewall you have.
    No matter on which interface this firewall rule is.

    2. The Upstream speed is equal to the speed of the fastest of the last interface up speed and the next to last interface up speed. As added to the load balance Pool.

    No.
    It's as fast as the upstream of the line on which you created the connection.

    3. The Downstream speed is equal to the sum of first interface down speed and the second interface down speed. As added to the load balance Pool.

    No.
    It's as fast as the downstream of the line on which you created the connection.

    The loadbalancer is in no way connected with the numbers of interfaces you have.
    Loadbalancing is just a special kind of policy routing.

    Since the loadbalancer is connection based: once a connection has been created, it will stay on the line.
    You never get a "sum of bandwidth".
    (unless you open up multiple connections, and each has a chance to land on another line).

    The pool order has absolutely no influence on what bandwidths you get.
    Are you sure you actually set the pool to loadbalance and not to failover?

    How did you do your tests?



  • Hi,

    Thanks for your comments.

    I will run another test that has all three interfaces load balanced and check the number of connections being setup between me and the speed test server. I think your explaination makes perfect sense.

    I can assure you that the pool was set to load balance and not failover. I have attached a spreadsheet that has the results of the test I did.

    Thanks

    Paul

    load_balancing.xls



  • The attached file is broken.
    (Not openable with OpenOffice Spreadsheet or Excel 2003)


Locked