Cisco: Port-channel load-balancing explanation [Part II]

As I promised in Part I of this article, here is the second part covering the port-channel load-balancing method explanation. If you didn’t know what I’m talking about, please be sure that you have read the first part of this article. Everything remains the same in this scenario. We have 3 physical interfaces bundled in one port-channel. Together with this port-channel we have some possible sources and destinations.

In this Part II, I will try to explain the remaining 6 methods of port-channel load-balance:

src-ip
src-mac
src-port
dst-mixed-ip-port
src-mixed-ip-port
src-dst-mixed-ip-port

src-ip / src-mac / src-port

I’ve grouped this 3 methods under one example as the basic principle is the same; Loads distribution on the source IP address / mac-address / port. Ignore completely the destination IP address / mac-address / port.

In the above case all traffic from Source A (depending on the method, this can be IP address A / mac-address A / port A)  is forward through physical interface Fa0/1 in Port-Channel 1, not matter of its destination. Fa0/2 and Fa0/3 are not an alternative in this load-balance methods.

dst-mixed-ip-port

Loads distribution on the destination IP address and TCP or UDP port

This method offer more granularity and we see that we start to have a more complex scenario as this process take into consideration a mix of IP address and TCP/ UDP port. We may have the following scenarios for traffic load-balance:

– from Src A packets to  Dst A and port 80 – Fa0/1 in PO1 – valid alternative
– from Src A packets to the same Dst A, but port 25 – Fa0/2 in PO1 – also valid, because the IP address is the same, but TCP port is different
– from Src A packes to Dst B, same port 25 – Fa0/2 in Po1 – valid as is the same port (25), but different IP address
– from Src A packets to Dst B port 25 – Fa0/3 in Po1 – not valid, as there is already a path through Fa0/2 for the packets matching this destination and port

src-mixed-ip-port

Loads distribution on the source IP address and the TCP or UDP port

This method offer also a great granularity in load-balance process. If we analyze the port communication trend, I would say that this method offer more granularity than the dst-mixed-ip-port one, because source ports are more random chosen in communication than destination port. Here we have the following scenarios:

– Src A and source port 32343 to Dst A – Fa0/1 in PO1 – valid choice
– Src A and source port 32345 to Dst A – Fa0/2 in PO1 – valid choice (same source IP, different source port)
– Src A and source port 32346 to Dst B – Fa0/2 in PO1 – valid choice (same source IP, different source port than previous example); you might think that also different destination, but in this method, destination IP or port are not taken into consideration.
– Src A and source port 32346 to Dst C – Fa0/3 in PO1 – not valid choice as the path for this source IP / source port is already defined through Fa0/2

src-dst-mixed-ip-port

Loads distribution on the source XOR-destination IP address and the TCP or UDP port

The best granularity until now. Almost every path in PO1 is a valid choice. You can just image that a path which is consider not valid is if you have a pair of  SRC IP : PORT -> DST IP: PORT and is already forwarded through one port (Fa0/2), then the Fa0/3 is not a valid choice for the same traffic. Otherwise, there are more possibilities to load balance traffic than in the previous methods. The issue is that not all devices support these last methods (especially the last 3), so if you’re device is capable to support this complex method you have to deal with the other ones and choose the best one for your scenario.

I want to close this article by adding a new load-balance method:

port-channel load-balance mpls

This method set the load-distribution method among the ports in the bundle for Multiprotocol Label Switching (MPLS) packets, use the port-channel load-balance mpls command in global configuration mode. I never had the chance to work with this method, so I don’t know how it’s working exactly (just in theory). If anybody has experience with it, I would be glad to add it’s explanation here (with credits of course).Otherwise you’ll have to wait until I get my hands on this configuration and then I’ll share my knowledge with you.