[c-nsp] Per Packet vs Per Destination Load Balancing

[Jay Hennigan]

On Fri, 22 Oct 2004, Chris Strandt wrote:

> Is there a performance impact or limitation with per destination load
> balancing?
>
> It seems to me that with per destination, IOS has to keep track of the
> source/destination pair.  Will this slow forwarding, or can this
> negativly impact a router forwarding a DDOS attack?

To make any routing decision, IOS needs to keep track of the source/
destination pair, or at a minimum the destination.  With per-destination
it must also keep track of the last such flow.  With per packet it must
also keep track of the link used by the last packet.  Assuming more than
one packet per flow, per-packet would result in a higher burden on the
router.  Some implementations result in these packets being process
switched, which is CPU-intensive.  CEF mitigates this.


The limitation of per-destination is that a given flow will not share
more than a single link.  The benefit is that packets will generally
arrive in order and jitter will be minimized.

The benefit of per-packet is even distribution of traffic and the
ability to utilize more than one link for a single flow, at the cost
of increased jitter and out-of-order packets.

During a distributed DoS attack, both methods will likely result in
similar impact.  In a non-distributed DoS attack, traffic will likely
saturate one link with per-destination and be distributed amongst the
links with per-packet.  The impact on performance during such an attack
would likely depend on a number of factors.

Some gory details:
http://www.cisco.com/en/US/tech/tk827/tk831/technologies_tech_note09186a0080094806.shtml

http://www.cisco.com/en/US/tech/tk365/tk80/technologies_tech_note09186a0080094820.shtml

--
Jay Hennigan - CCIE #7880 - Network Administration - jay at west.net
WestNet:  Connecting you to the planet.  805 884-6323      WB6RDV
NetLojix Communications, Inc.  -  http://www.netlojix.com/