Regarding ECMP hashing

Question

Assume we have a following topology. R1 represents Router1, R2 represents Router2, R3 represents Router3 and R4 represents Router4.

Src represents the source Server.

Dest represents the destination.

There are 3 distinct paths to reach from Source server to the destnation.

R1--R2--R4 is path1, R1--R4 is path2, R1--R3--R4 is path3.

Assume, all of these are equal cost paths. BGP is running as the routing protocol and maximum paths is enabled for BGP.

Hence, we will have all the 3 paths to reach from R1 to R4.To be more precise, R1 will have all the 3 paths installed in the routing table, to reach R4.

Question1:

Let's say there is an application (which uses TCP) running on Src on port number 6811 and performing a transaction to destination on port number 53 for a period of 30 minutes.

So calculating the hash based on Source IP, Destination IP, Source Port, Destination Port and Protocol,we would get a unique hash value and this would be a TCP flow.

I believe that out of the 3 distinct paths mentioned above, only 1 of the paths will be used (even though we have BGP ECMP enabled). The remaining 2 paths will be unused since the hash value remains unchanged for those 30 minutes.

Is the above understanding correct ?

Question 2: Let's say there is another application (which uses TCP) running on Src on port number 6812 and performing a transaction to destination on port number 64 for a period of 10 minutes. (Note: source port and destination port have changed as compared to Question 1 scenario). So calculating the hash based on Source IP, Destination IP, Source Port, Destination Port and Protocol. We would get a unique hash value and this would be a different TCP flow. This hash value will definitely be different from the one calculated in Question1.

I believe that out of the 3 distinct paths mentioned above, only 1 of the paths will be used (even though we have BGP ECMP enabled). The remaining 2 paths will be unused since the hash value remains unchanged for those 10 minutes.

Is the above understanding correct ?

:) not, homework, i created the scenario on my own and trying to get clarity. — fsociety, Dec 30 '18 at 13:25
ECMP stands for "*equal-cost* multipath routing" - you've not mentioned the route costs at all... — Zac67, Dec 30 '18 at 13:28
Paths or links of equal cost? *Paths* is highly unlikely, being of unequal lengths. — Zac67, Dec 30 '18 at 16:33
@Zac67 i meant to say that R1 has three equal cost paths in its routing table to reach R4 — fsociety, Dec 30 '18 at 16:39

Ron Trunk · Accepted Answer · 2018-12-30T14:59:42.490

3

Yes, you are correct on both questions. To make it simpler: ECMP and its hashing algorithm selects a path per flow. So, a single flow will always use the same path *.

While the aggregate bandwidth of ECMP is the sum of all the links, a single flow cannot exceed the bandwidth of a single link.

* Per-packet ECMP is available in some scenarios, but is almost never used in actual networks.

edited Dec 30 '18 at 14:59

answered Dec 30 '18 at 13:59

Ron Trunk

66,852
5
65
126

One question here. In an ECMP path(consider 2 path ECMP), if one of the path(say path-1) goes down, all the flows will be mapped the remaining one path(say path2). Now, after a while, if the path1 comes back up, how will the load balancing happen ? Will some the flows will be forcefully re-mapped to the recovered path ? – Hemanth May 30 '22 at 19:43

Regarding ECMP hashing

1 Answers1

Linked