White Papers

Polarization
Multipath routing is a method that is often used to address data forwarding issues during network failures so that the network
traffic reaches its desired destination. Multipath routing in IP networks is typically implemented using Equal-Cost Multipath
(ECMP) routing, which employs load balancing algorithms to distribute the traffic over multiple paths towards its destination.
In a multi-tier network where load balancing is performed at each tier, static hash algorithms polarize the traffic where load
balancing is ineffective in the higher tiers. The polarization effect is exaggerated if all the nodes in the network have to choose
from the same set of ECMP paths. Traffic polarization results in packet reordering and route flapping. The following figure
explains the traffic polarization effect. Router B performs the same hash as router A and all the traffic goes through the same
path to router D, while no traffic is redirected to router E. The following figure explains the traffic polarization effect:
Figure 39. Before Polarization Effect
Router B performs the same hash as router A and all the traffic goes through the same path to router D, while no traffic is
redirected to router E.
Some of the anti-polarization techniques used generally to mitigate unequal traffic distribution in LAG/ECMP as follows:
1. Configuring different hash-seed values at each node - Hash seed is the primary parameter in hash computations that
determine distribution of traffic among the ECMP paths. The ECMP path can be configured different in each of the nodes
“hash-algorithm seed-value” would result in better traffic distribution for a given flow, by reducing Polarization effect.
2. Configuring Ingress port as an additional load-balancing parameters [using “load-balance ingress-port enable”] would reduce
the polarization effect.
3. Configuring different load-balancing parameters at each tier. In Router A, the hash fields for load balancing could be
source-ip, dest-ip, vlan, protocol, L4-source-port and L4-dest-port, whereas on Router B, the hash fields use only source-ip,
dest-ip, and protocol
4. Configuring different hash algorithms at different tiers. For example, Router A could use crc16 as the hash algorithm while
router B can use XOR16 as the hash algorithm.
Configuration and Benefits
The preceding anti-polarization techniques require some coordinated configuration of network nodes to solve the problem and
these techniques are not scalable when the number of tiers in the network is high. Flow based hashing specifically addresses
this using Macro flow-based Hash function. It facilitates a dynamic hash function selection across different nodes in a network
on a macro flow basis, thus reducing unfair distribution of bandwidth between members and starvation.
Equal Cost Multi-Path (ECMP)
317