ECMP Load-Balancing Algorithms

Hash-based algorithms prioritize session stickiness
—The
IP Modulo
and
IP Hash
algorithms use hashes based on information in the packet header, such as source and destination address. Because the header of each flow in a given session contains the same source and destination information, these options prioritize session stickiness. If you choose the
IP Hash
algorithm, the hash can be based on the source and destination addresses, or the hash can be based on the source address only. Using an IP hash based on only the source address causes all sessions belonging to the same source IP address to always take the same path from available multiple paths. Thus the path is considered sticky and is easier to troubleshoot if necessary. You can optionally set a
Hash Seed
value to further randomize load balancing if you have a large number of sessions to the same destination and they’re not being distributed evenly over the ECMP links.
Balanced algorithm prioritizes load balancing
—The
Balanced Round Robin
algorithm distributes incoming sessions equally across the links, favoring load balancing over session stickiness. (Round robin indicates a sequence in which the least recently chosen item is chosen.) In addition, if new routes are added or removed from an ECMP group (for example if a path in the group goes down), the virtual router will re-balance the sessions across links in the group. Additionally, if the flows in a session have to switch routes due to an outage, when the original route associated with the session becomes available again, the flows in the session will revert to the original route when the virtual router once again re-balances the load.
Weighted algorithm prioritizes link capacity and/or speed
—As an extension to the ECMP protocol standard, the Palo Alto Networks
®
implementation provides for a
Weighted Round Robin
load-balancing option that takes into account differing link capacities and speeds on the egress interfaces of the firewall. With this option, you can assign
ECMP Weights
(range is 1 to 255; default is 100) to the interfaces based on link performance using factors such as link capacity, speed, and latency to ensure that loads are balanced to fully leverage the available links.
For example, suppose the firewall has redundant links to an ISP: ethernet1/1 (100 Mbps) and ethernet1/8 (200 Mbps). Although these are equal-cost paths, the link via ethernet1/8 provides greater bandwidth and therefore can handle a greater load than the ethernet1/1 link. Therefore, to ensure that the load-balancing functionality takes into account link capacity and speed, you might assign ethernet1/8 a weight of 200 and ethernet1/1 a weight of 100. The 2:1 weight ratio causes the virtual router to send twice as many sessions to ethernet1/8 as it sends to ethernet1/1. However, because the ECMP protocol is inherently session-based, when using the
Weighted Round Robin
algorithm, the firewall will be able to load balance across the ECMP links only on a best-effort basis.
Keep in mind that ECMP weights are assigned to interfaces to determine load balancing (to influence which equal-cost path is chosen), not for route selection (a route choice from routes that could have different costs).
Assign lower-speed or lower-capacity links with a lower weight. Assign higher-speed or higher-capacity links with a higher weight. In this manner, the firewall can distribute sessions based on these ratios, rather than overdrive a low-capacity link that is one of the equal-cost paths.