What is the correct xOS address-based algorithm for VMWare ESXi?

[Anonymous]

I am migrating from a Dell switch stack to a stack of X440 switches. I have 3 VMware ESXi hosts to attach to the stack; each host uses a link aggregate of 6 physical connections.

VMware's knowledge base states that ESXI only supports an "IP HASH" load sharing algorithm. This is a Cisco term that does not appear in the xOS documentation.

My SummitStack will support address-based load sharing, but what is the correct algorithm for ESXi?

You can use the default (L2) but then ESXi won't perform the load sharing, which defeats the purpose of having a link aggregate.

The Dell switch stack uses "Source IP address and source port". There is no corresponding xOS algorithm.

L3 uses both the source and destination IP Addresses; L3_L4 uses the source and destination IP addresses and ports.

[Anonymous]

Anyway, after making the configuration changes, and putting some load on the ESXi host, the utilization statistics look encouraging:

Slot-1 Stack.1 # show port A,B,C,D,E,F utilization
Link Utilization Averages Mon Sep 28 15:37:26 2015
Port Link Rx Peak Rx Tx Peak Tx
State bytes/sec bytes/sec bytes/sec bytes/sec
================================================================================
A A 0 556377 53318556 53318556
B A 0 143009 70 128803140
C A 0 3209 11 130255665
D A 521464 845687 105 64082
E A 11 9572 64616223 64616223
F A 175927 2129075 94 1803197Since the algorithm uses both source and destination, transmits and receives will have different hashes and may go across different member ports. Since there are 3 VLANs, it's possible that all 6 (=3x2) member ports will be used.

[Patrick_Voss]

I just wanted to make sure. It is a common misconception .

[Anonymous]

Yes, load sharing is not load balancing. But it's better than everything going across a single wire.

[Patrick_Voss]

Yes, the utilization shows ingress and egress traffic. You can toggle different outputs using the space bar. Please keep in mind that this output is an average and that a lag will never evenly distribute information across both links.