The Load Balancer in Azure, is for many of us, a black box where things happens without our knowing. Here is some information that will open up this box:
(I have stolen all this information from this blog post.)
1. The distribution of the load is not round robin
Microsoft Azure Load Balancer is a Layer-4 type load balancer. Microsoft Azure load balancer distributes load among a set of available servers (virtual machines) by computing a hash function on the traffic received on a given input endpoint. The hash function is computed such that all the packets from the same connection (TCP or UDP) end up on the same server. The Microsoft Azure Load Balancer uses a 5 tuple (source IP, source port, destination IP, destination port, protocol type) to calculate the hash that is used to map traffic to the available servers. The hash function is chosen such that the distribution of connections to servers is fairly random. However, depending on traffic pattern, it is possible for different connections to get mapped to the same server. (Note that the distribution of connections to servers is NOT round robin, neither there is any queuing of requests, as has been mistakenly mentioned in some articles and some blogs). The basic premise of the hash function is given a lot of requests coming from a lot of different clients, you will get a nice distribution of requests across the servers.
2. Intra Data Center Traffic Optimization
Microsoft Azure Load Balancer optimizes traffic between Microsoft Azure datacenters in the same region, such that traffic between Azure tenants that talk over the VIP and are within the same region, after TCP/IP connection is initiated, they bypass Microsoft Azure Load Balancer altogether.