Re: Problem with etherchannel between 2 3550 FXs

In article <Xns96C394B8549CAqubepunkasscom@xxxxxxxxx>,
Who knows? <qube99@xxxxxxxxx> wrote:
:I have a 3550's in different buildings with ~2km of multimode fiber between
:them. Until we upgrade to singlemode, we want to use etherchannel to get us
:to at least 400mbps.

:The link works great, but can't seem to get more than 100mbps over the
:link.

In the configuration portion you included, you did not have any
command explicitly setting the mechanism for a selecting particular
link for any one connection.

For any constant number of active links, the choice of links for
any one connection is deterministic, based upon the source
and destination MAC addresses. Essentially it takes some number
of the final bits of each of the source and destination
MACs, performs a logical operation on them (e.g., XOR), and
the numeric result selects which of the links is used.

You get to choose which logical operation you want, but for any
one source + destination pair, the result is going to be constant
within the chosen operation.

The implication of this is that if your traffic is all between
two fixed MACs (e.g., if there is a routing interface on
either side of the link) then the traffic would all go over
one link. Or if the sources differ but the destination is
the same each time, and you happen to have chosen the logical
operation that ignores the source MAC, then it'll end up going
over the same link...

EtherChannel is not really designed for "load-balancing" as such:
it is more a way to allow multiple connections (between different
systems) in parallel.


If I recall correctly, you can also select per-packet load balancing,
but then you run the risk that packets will arrive out of order,
which might matter very little or might matter a lot in some cases.
(e.g., for TCP with Selective ACKs turned on, it might be fine,
provided you aren't fragmenting packets; but for multimedia streaming
out-of-order packets can be big trouble.)
--
Any sufficiently advanced bug is indistinguishable from a feature.
-- Rich Kulawiec
.

Relevant Pages

  • Re: Problem with etherchannel between 2 3550 FXs
    ... > and destination MAC addresses. ... > of the final bits of each of the source and destination ... > You get to choose which logical operation you want, ... > provided you aren't fragmenting packets; ...
    (comp.dcom.sys.cisco)
  • Re: ARP requests on my net?
    ... But my machine is not configured to be a router, bridge, or gateway. ... But why would it need the MAC for the router? ... my machine is the destination IP. ... only for passing along to another device closer to the destination ...
    (Fedora)
  • Re: Ip forwarding
    ... Packets that are meant to be forwarded by a router must be sent to its MAC address on the link. ... The gateway column optionnally contains the address of the next hop router if the destination is not directly reachable on the network attached to the output interface. ... In both cases the packet is sent on the link to the next hop MAC address. ...
    (comp.os.linux.networking)
  • Re: Port Mirroring in Linux
    ... will be the MAC of other NIC), similar to the DNAT in iptables. ... But why redirect only IPv4 traffic? ... one you wanted to redirect your traffic to or simply the sniffer interface. ... destination, this command will alter the destination. ...
    (comp.os.linux.networking)
  • Re: Hey Jim! Need to find an article on "isaserver.org" but can find it
    ... It was the MAC address that became the problem, ... source MAC and the destination MAC were identical in the packet header ... rather than just going directly to the resource as they should have done. ... listed in both the Source and Destination field of the Layer2 Header ...
    (microsoft.public.isa)