PIM ECMP Load Splitting

The PIM ECMP feature allows downstream PIM routers to choose multiple ECMP paths to source via hash from one of following selections without affecting existing unicast routing algorithm:

This feature operates on a per (S,G) basis splitting the load onto available equal-cost paths by hashing according to the selection criteria configured by the user. It does not operate by counting the flows. Load splitting need not balance the traffic on the available paths. PIM ECMP load splitting uses a hash algorithm based on the selected criteria to pick up the path to use and will result in load-sharing the traffic when there are many multicast streams that utilize approximately the same amount of bandwidth.

PIM ECMP Load Splitting Based on Source Address

When you enable PIM ECMP load splitting based on source address, the RPF interface for each (*, G) or (S,G) state is selected among the equal cost paths based on the hash derived from the source address. For an (S, G) state, the address considered for hashing is the source address of the state. For a (*, G) state, the address considered for hashing is the address of the RP that is associated with the state‘s group address. There is no randomization applied when calculating the hash value. The same hash value is generated on all the ExtremeXOS routers for a given source address. If there are two equal cost paths ("left" and "right") available at the last hop router and at each of the intermediate routers for a given source, each of these routers pick the same hash, and the traffic flows can get skewed (to either "left" or "right" paths).

PIM ECMP Load Splitting Based on Group Address

When you enable PIM ECMP load splitting based on group address, the RPF interface for each (*, G) or (S,G) state is selected among the equal cost paths based on the hash derived from the group address. If multiple equal cost common paths exist to the multicast source and the RP that is associated with the state‘s group address, the same hash will be chosen for both (*, G) and (S, G) states as the same group address is used in deriving the hash. There is no randomization applied when calculating the hash value. The same hash value is generated on all the ExtremeXOS routers for a given group address. If there are two equal cost paths ("left" and "right") available at the last hop router and at each of the intermediate routers for a given group, each of these routers pick the same hash and the traffic flows can get skewed (to either "left" or "right" paths).

PIM ECMP Load Splitting Based on Source-Group Addresses

When you enable PIM ECMP load splitting based on source-group address, the RPF interface for each (*, G) or (S,G) state is selected among the equal cost paths based on the hash derived from the source and group addresses. For an (S, G) state, the address considered for hashing is the source address of the state. For a (*, G) state, the address considered for hashing is the address of the RP that is associated with the state‘s group address. There is no randomization applied when calculating the hash value. The same hash value is generated on all the ExtremeXOS routers for a given source-group address. If there are two equal cost paths ("left" and "right") available at the last hop router and at each of the intermediate routers for a given source-group, each of these routers pick the same hash and the traffic flows can get skewed (to either "left" or "right" paths).

PIM ECMP Load Splitting Based on Source-Group-Next Hop Addresses

When you enable PIM ECMP load splitting based on source-group-next hop address, the RPF interface for each (*, G) or (S,G) state is selected among the equal cost paths based on the hash derived from the source, group and next hop addresses. The hash value derived after introducing the next hop address is still predictable as there is no randomization applied when calculating the hash value. However, since the next hop address used at each of the routers vary, the hash value generated on each of the ExtremeXOS routers is different. As the hash value is different on each of the routers, the problem of traffic path skew present in the above mentioned schemes does not exist in this scheme.

Reconvergence Due to Unicast Routing Changes

When a unicast route to a source or RP address changes (when a path goes down or a new path becomes available), all the (*, G) and (S, G) states change based on the available unicast route information provided by Route Manager process. If one of the paths goes down and comes back up, multicast forwarding will reconverge to same RPF path that was used before the path went down. The hash function based on Source-Group-Next Hop avoids skewing of traffic flows because it introduces the actual next-hop IP address of PIM neighbors into the calculation resulting in different hash value being computed for each router. The Source-Group-Next Hop based hash function doesn‘t take the total number of available paths into consideration and so it increases stability of the paths chosen during path failures. During path failures, the multicast states that were using the failed path would need to reconverge onto the remaining paths. All other states using the unaffected paths are not affected.

Limitations