[j-nsp] 802.3ad Question

alexi acronopio at gmail.com
Thu Jan 14 00:57:35 EST 2010


Hello Stefan:

Thanks for your answer, this was a costumer question that i think camed from
the way how load balancing in Cisco using Ether Channel was done , in that
case you had a fixed number of 8 values so the hashing algorithm (using the
source ip , mac ... etc) gives you one of those fixed values. So each one of
the links in a bundle was identified with one or several of those values ...
so in this case you can only have perfect load balance if you are using 2, 4
or 8 links ... any other combination gives you a not even distribution ...

this is explained here :
http://www.cisco.com/en/US/tech/tk389/tk213/technologies_tech_note09186a0080094714.shtml

I don´t know how this goes in 802.3ad  in Juniper ...what do you think ? ,

have you ever seen examples of good load balancing using 3 , 5 or 6 links in
a bundle ?

BR
Alexi


On Thu, Jan 14, 2010 at 2:04 AM, Stefan Fouant <
sfouant at shortestpathfirst.net> wrote:

> > -----Original Message-----
> > From: juniper-nsp-bounces at puck.nether.net [mailto:juniper-nsp-
> > bounces at puck.nether.net] On Behalf Of alexi
> > Sent: Wednesday, January 13, 2010 11:45 PM
> > To: mtinka at globaltransit.net
> > Cc: juniper-nsp at puck.nether.net
> > Subject: Re: [j-nsp] 802.3ad Question
> >
> > Hello Mark:
> >
> > I am coming back to this question , i would apreciate your help again
> > ...
> >
> > in the example you mention you said that your bundle is using 3xGE and
> > you
> > have a pretty fair load balance 1:1:1
> > do you know if there is any requirement about having pair numbers or
> > power
> > of 2 amount of links in a bundle to get a good load balance ?
> >
> > I guess that should depend on the hashing algorithm but of course there
> > is
> > no much info about how it works .... we are interested to have 6xGE in
> > a
> > bundle ... so , will we get 6Gbps of real throuhput  ...?
>
> The hashing algorithm Juniper uses is proprietary so there isn't much
> useable information out that, but in my experience I've never seen anything
> along the lines which would require a LAG to be comprised of multiples of 2
> links to get an even load balance.  The load balancing tends to get a more
> even distribution when you've got a large number of flows.  Assuming a
> large
> number of flows, you should be able to get an even balance across the
> component links within the 802.3ad LAG bundle.  You might also want to
> consider enabling Layer 4 hashing to get even more granular distribution,
> if
> the source and destinations are sparse, but you're using a large number of
> source and destination ports.
>
> HTHs.
>
> Stefan Fouant, CISSP, JNCIE-M/T
> www.shortestpathfirst.net
> GPG Key ID: 0xB5E3803D
>
>


More information about the juniper-nsp mailing list