Re: Internet Traffic load balancing

Marko,
I made a small test and the thing makes for a lot of fun (and food for
thought for a R/S expert :)

I have one generator (ostinato), sending steady traffic to 10 different
addresses, 1.1.1.1 .. 10.10.10.10 (@1000 pps) to a router R1 which has
2 serials connected to router R2 acting as a sink.
(Actually sending pings, so the sink is answering back)

Serial links are 20.1.1.0/30 and 20.2.2.0/30, generated 10 statics to
0.0.0.0/0 via 169.254.1.1 .. 7 and 169.254.2.1 .. 3, and respective
recursive routes to s0/0 and s0/1.

SNMP in R1 shows more or less balanced traffic :(
Hmm, I got 2 routes out (so ended with 8:1) and all traffic went to one
interface! What ? Added a second route and I got a 5:2 relation.
Aha. The 10 different destinations are not getting into 10 different
CEF hash buckets!

R1 is a 2610XM running 12.4.9T7.
I played with "ip cef load-sharing algorithm" and the load changes.

So my readings:

works ? yes. (unequal load balancing using CEF, seeded by statics)
is obvious how ? Not at all.

This version does not has 16 fixed buckets as Ivan documents:

R1#sh ip cef 0.0.0.0 0.0.0.0 internal
*Jul 1 12:30:36.866: %SYS-5-CONFIG_I: Configured from console by console
0.0.0.0/0, version 50, epoch 0, per-destination sharing
0 packets, 0 bytes
   via 169.254.2.2, 0 dependencies, recursive
     traffic share 1
     next hop 20.2.2.2, Serial0/1 via 169.254.2.2/32
     valid adjacency
   via 169.254.2.1, 0 dependencies, recursive
     traffic share 1
     next hop 20.2.2.2, Serial0/1 via 169.254.2.1/32
     valid adjacency
   via 169.254.1.7, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.7/32
     valid adjacency
   via 169.254.1.6, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.6/32
     valid adjacency
   via 169.254.1.5, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.5/32
     valid adjacency
   via 169.254.1.4, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.4/32
     valid adjacency
   via 169.254.1.3, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.3/32
     valid adjacency
   via 169.254.1.2, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.2/32
     valid adjacency
   via 169.254.1.1, 0 dependencies, recursive
     traffic share 1
     next hop 20.1.1.2, Serial0/0 via 169.254.1.1/32
     valid adjacency

   0 packets, 0 bytes switched through the prefix
   tmstats: external 0 packets, 0 bytes
            internal 0 packets, 0 bytes
   Load distribution: 0 1 2 3 4 5 6 7 8 (refcount 1)

   Hash OK Interface Address Packets
   1 Y Serial0/1 point2point 0
   2 Y Serial0/1 point2point 0
   3 Y Serial0/0 point2point 0
   4 Y Serial0/0 point2point 0
   5 Y Serial0/0 point2point 0
   6 Y Serial0/0 point2point 0
   7 Y Serial0/0 point2point 0
   8 Y Serial0/0 point2point 0
   9 Y Serial0/0 point2point 0
   refcount 4318, covered prefixes:
     169.254.2.3/32
     169.254.2.2/32
     169.254.2.1/32
     169.254.1.7/32
     169.254.1.6/32
     169.254.1.5/32
     169.254.1.4/32
     169.254.1.3/32
     169.254.1.2/32
     169.254.1.1/32
     172.30.0.0/16
     20.2.2.0/30
     20.1.1.0/30
     224.0.0.0/4
     255.255.255.255/32
     0.0.0.0/32
R1#

-Carlos

Marko Milivojevic @ 02/05/2011 22:39 -0300 dixit:
> On Mon, May 2, 2011 at 17:45, Carlos G Mendioroz <tron_at_huapi.ba.ar> wrote:
>> 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.
>
> Those are valid points, but are contrary to what was originally
> expected. My argument was that CEF would break this, but I honestly
> expected that it would work when process-switched. It doesn't seem to.
> I will add 3rd router behind R2 and re-run the test.
>
> --
> 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
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
>
>
>
>
>
>
>

-- 
Carlos G Mendioroz  <tron_at_huapi.ba.ar>  LW7 EQI  Argentina
Blogs and organic groups at http://www.ccie.net