Re: PAgP Learn Method and Port-channel Loadbalancing

Well, if we follow the same line of thought of packets flowing one
direction, the other direction (where the src add is now dest add) would
have a hash as well and could fall on any interface of the bundle... right?

On Tue, Nov 10, 2009 at 8:03 PM, S Malik <ccie.09_at_gmail.com> wrote:

> I think , what should happen is that traffic in both directions should use
> the same interface (just a guess).
>
>
> On Tue, Nov 10, 2009 at 4:30 PM, Miroslav Kosut <miroslav.kosut_at_me.com
> >wrote:
>
> > Yes correct, we are getting closer. My question is weird, I know :-)
> >
> > We LOAD-BALANCE outgoing frames and we LEARN from incoming frames.
> > Completely true.
> >
> > However,
> > what If I learned the MAC address on the physical interface from incoming
> > frame, and I would like to send a frame in the opposite direction (a
> reply
> > to the previous frame, sent to the same host). Will I use the MAC address
> > entry that has been learned a while before? I think yes, but I am not
> sure.
> > I simply think that the load balancing will not occur when the MAC is
> > located behind the physical interface (instead of aggregate interface -
> > Po1).
> >
> > Regards,
> > Miroslav Kosut
> >
> > On Nov 10, 2009, at 10:19 PM, S Malik wrote:
> >
> > In my opinion (I could be wrong) pagp learn-method is setting the
> > source-address learning of incoming packets received from an
> ether-channel,
> > now as far as LB in ether-channels is concerned, it is done for the
> outgoing
> > packets not incoming.
> >
> >
> > On Tue, Nov 10, 2009 at 4:00 PM, Miroslav Kosut
> <miroslav.kosut_at_me.com>wrote:
> >
> >> Assume the following scenario:
> >>
> >> Two switches (SW1 and SW2), interconnected via 4-membered L2
> etherchannels
> >> (on both ends, using fa0/1 - 4).
> >>
> >> SW1 fa0/1 <------------> SW2 fa0/1
> >> SW1 fa0/2 <------------> SW2 fa0/2
> >> SW1 fa0/3 <------------> SW2 fa0/3
> >> SW1 fa0/4 <------------> SW2 fa0/4
> >>
> >> Each physical interface on SW1 is configured with the command:
> >>
> >> *pagp learn-method physical-port*
> >>
> >> Then, some frame with src mac address DE:AD:DE:AD:DE:AD :-) is received
> on
> >> the SW1 etherchannel, on the fa0/2 physically. We are using
> physical-port
> >> learning so the mac address entry looks like this:
> >>
> >> *DE:AD:DE:AD:DE:AD fa0/2 (if I was using "aggregate-port" learning
> >> method, I would see here, let's say Po1)*
> >>
> >> If SW1 receives the frame with a destination mac address of
> >> DE:AD:DE:AD:DE:AD, will it load-balance using a hash? It is already
> learned
> >> on the physical interface so I don't see any reason of wasting a time
> with
> >> loadbalancing.
> >>
> >> If my question is completely wrong, let me know.
> >>
> >> Regards,
> >> Miroslav
> >>
> >> On Nov 10, 2009, at 9:39 PM, Rick Mur wrote:
> >>
> >> The loadsharing on portchannels works with so-called hashes. In your
> >> case for each destination MAC address that is seen outgoing on the
> >> port-channel a hash is generated and divided over 1 of the links
> (buckets).
> >> So after a while all the possible variations are divided and are
> 'shared'
> on
> >> the links. But keep in mind that this is fixed, traffic to the same
> >> destination MAC address is using only 1 of the links.
> >>
> >> Therefore it's highly recommended to use layer 3 and if possible layer 4
> >> information to get the best balancing over the links in the channel.
> >>
> >>
> >> --
> >> Regards,
> >>
> >> Rick Mur
> >> CCIE2 #21946 (R&S / Service Provider)
> >> Sr. Support Engineer IPexpert, Inc.
> >> URL: http://www.IPexpert.com <http://www.ipexpert.com/> <
> http://www.ipexpert.com/>
> >>
> >> On 10 nov 2009, at 21:09, Miroslav Kosut wrote:
> >>
> >> Dear group,
> >>
> >>
> >> a quick question to confirm my understanding. Suppose I configure the
> >> following on the same etherchannel:
> >>
> >>
> >> 1. PAgP learning method to be a "physical-port learning"
> >>
> >> 2. Port-channel loadbalancing to be based on a "destination mac address"
> >>
> >>
> >> Then, am I right when I am saying that load-balancing method is ignored
> >> when a destination mac address is already learned (on a physical-port) ?
> >>
> >>
> >> My understanding:
> >>
> >> Since the mac is already learned on a physical-port, there is no need to
> >> calculate the outgoing member interface (load-balancing).
> >>
> >> Of course, if the mac address hasn't been learned, frames would be
> >> unicast-flooded. Then, load balancing IS performed and the outgoing
> >> interface is chosen.
> >>
> >>
> >> Regards,
> >>
> >> Miroslav Kosut
> >>
> >>
> >>
> >> Blogs and organic groups at http://www.ccie.net
> >>
> >>
> >> _______________________________________________________________________
> >>
> >> Subscription information may be found at:
> >> http://www.groupstudy.com/list/CCIELab.html
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html

Blogs and organic groups at http://www.ccie.net
Received on Thu Nov 12 2009 - 15:24:26 ART