Re: Internet Traffic load balancing

Can I opt for unflavored water? ;-)

--
Marko Milivojevic - CCIE #18427
Senior Technical Instructor - IPexpert
Free CCIE Training: http://bit.ly/vLecture
Mailto: markom_at_ipexpert.com
Telephone: +1.810.326.1444
Community: http://www.ipexpert.com/communities
:: Sent from my phone. Apologies for errors and brevity. ::
On May 2, 2011, at 18:48, ALL From_NJ <all.from.nj_at_gmail.com> wrote:
> When we get to Vegas, I will buy / bring you a free Coors or Bud at the
booth at the solutions center.
>
> LOL
>
>
> .
> On Mon, May 2, 2011 at 9:41 PM, Marko Milivojevic <markom_at_ipexpert.com>
wrote:
> I'd love to, but that, once again, requires packet generator, which I
> don't have readily available. Someone else needs to volunteer for this
> one :-). I'll keep on slamming the CEF solution, as I'm curious about
> it. PBR... nah, never scales - it's a band-aid.
>
> --
> Marko Milivojevic - CCIE #18427
> Senior Technical Instructor - IPexpert
>
> FREE CCIE training: http://bit.ly/vLecture
>
> Mailto: markom_at_ipexpert.com
> Telephone: +1.810.326.1444
> Web: http://www.ipexpert.com/
>
> On Mon, May 2, 2011 at 18:39, ALL From_NJ <all.from.nj_at_gmail.com> wrote:
> > Very cool to see this idea, wow ... I would not have thought of that.
> >
> > I would however, be more inclined to consider a different solution.  The
CEF
> > load balancing idea is fine if all flows are equal ... but ... as we
know,
> > no all flows are equal.  And ... nothing will come out even or exactly
> > 70/30% either ... so we should probably aim for 'close-enough'.
> >
> > What type of traffic are we talking about?
> >
> > What about the millisecond bursts, what if a 'heavy' flow goes out the
> > slower link?  What if a voice and or video etc ... also happen to also go
> > out the slow link?  Which flow gets dropped, delayed, shaped ... etc
....?
> > Is this ok?
> >
> > Since this clever solution only considers L3, you may be causing more
harm
> > at various times in your network.
> >
> > From a production standpoint, this would be hard to pinpoint and tshoot
> > since the nature of flows is changing ... good today, bad tomorrow, good
at
> > 10 am, but not at 1PM when people come back from lunch ... etc ...
> >
> > Might be best to look at PBR or another solution which can distribute the
> > load based on packet sizes or protocol.  Maybe send just http or
something
> > similar out the slow link, and send everything else out the larger
> > connection.
> >
> > Also, whatever solution is chosen, how will you know if you have
configured
> > the right solution?   You will need to monitor the interfaces / queues
for
> > drops etc ...
> >
> > I am learning from you all!  Nice to see this thread!!!   You guys
totally
> > rock.
> >
> > I vote for a lab test ... who can test this, PBR, and this scenario?
> > "Inquiring minds what to know" ... as the expression goes.
> >
> > All those in favor of Marko testing this, say "Aye" .  I think the group
> > votes for you Marko ...
> >
> > ;-)
> >
> > .
> > On Mon, May 2, 2011 at 8:55 PM, <ron.wilkerson_at_gmail.com> wrote:
> >>
> >> Most definitely per packet.
> >> Sent from my Verizon Wireless BlackBerry
> >>
> >> -----Original Message-----
> >> From: Carlos G Mendioroz <tron_at_huapi.ba.ar>
> >> Sender: nobody_at_groupstudy.com
> >> Date: Mon, 02 May 2011 21:45:14
> >> To: Marko Milivojevic<markom_at_ipexpert.com>
> >> Reply-To: Carlos G Mendioroz <tron_at_huapi.ba.ar>
> >> Cc: Brian McGahan<bmcgahan_at_ine.com>; masroor ali<masror.ali_at_gmail.com>;
> >> Cisco certification<ccielab_at_groupstudy.com>
> >> Subject: Re: Internet Traffic load balancing
> >>
> >> This was generated from the switching point ? I.e. it was generated by
> >> the router having the 5:2 statics ?
> >>
> >> I would test it from a neighbour, so CEF is used for sure. Local trafic
> >> is process switched most probably.
> >>
> >> -Carlos
> >>
> >> Marko Milivojevic @ 02/05/2011 21:10 -0300 dixit:
> >> >> I wonder if we could test this with a simple ping?
> >> >
> >> > Disclaimer: This is unbelievably unscientific, quick and dirty and in
> >> > no way proves either point. I still think a proper traffic generator
> >> > is required.
> >> >
> >> > I added 7 loopbacks on another router (the one that is default gateway
> >> > for both interfaces):
> >> >
> >> > 101.100.100.100
> >> > 102.100.100.100
> >> > 103.100.100.100
> >> > 104.100.100.100
> >> > 105.100.100.100
> >> > 106.100.100.100
> >> > 107.100.100.100
> >> >
> >> > I ran the following from the router with static routes (prior to this,
> >> > I made sure the ARP table was populated and I cleared the counters and
> >> > disabled anything else on the router than can generate packets and/or
> >> > frames, leaving only my pings in output counters):
> >> >
> >> > foreach ip {
> >> >  101.100.100.100
> >> >  102.100.100.100
> >> >  103.100.100.100
> >> >  104.100.100.100
> >> >  105.100.100.100
> >> >  106.100.100.100
> >> >  107.100.100.100
> >> > } { ping $ip repe 1 }
> >> >
> >> > What I should be seeing is 5:2 ratio in packets. This is what I got:
> >> >
> >> > R2#sh int gi0/0 | i packets out
> >> >      4 packets output, 456 bytes, 0 underruns
> >> > R2#sh int gi0/1 | i packets out
> >> >      3 packets output, 342 bytes, 0 underruns
> >> >
> >> > Which leads me to my original assumption of 1:1 ratio.
> >> >
> >> > --
> >> > Marko Milivojevic - CCIE #18427
> >> > Senior Technical Instructor - IPexpert
> >> >
> >> > FREE CCIE training: http://bit.ly/vLecture
> >> >
> >> > Mailto: markom_at_ipexpert.com
> >> > Telephone: +1.810.326.1444
> >> > Web: http://www.ipexpert.com/
> >> >
> >> >
> >> > Blogs and organic groups at http://www.ccie.net
> >> >
> >> >