Load Balancing

What is Load Balancing ?

Distributing network traffic across multiple servers effectively is called Load Balancing.To meet high volumes, it generally requires adding more servers. Load Balancer routes the client requests to the servers in an optimized way and takes care that no server is overworked. If a new server is added to the group, It will start sending requests to this new server.

Load-Balancing Algorithms

Why did algorithms come into the picture ? Actually Load Balancer selects the server based on two factors,

periodically pinging server to check its availability.
defined algorithms - based on which load-balancer selects the server.

Algo 1 Least Connection (Default algorithm)

The least connection is the default algorithm to select server. Server which has the least number of active transactions is picked by Load Balancer to handle request. Load Balancer maintains records of transactions for each server.

For example,

See the below figure, Load-Balancer will select server 2.

Now discuss this in detail: Let's say all the servers have the same capacity and still some of the servers are overloaded as there may be the situation that the client stayed in those servers for longer duration and connected to other severs for shorter duration. Active connections on the server where client stayed longer will pile up and hence based on the least connection algorithm Load Balancer will route the request in the server with least active connections,

Weighted Least Connections

Now let's consider the above Least Connection algorithm with different server capacities (CPU/RAM etc.). You would definitely want to allocate more requests to the higher capacity server than the lower capacity servers. Weighted Least Connection is the solution.

Let's say there are 2 servers, server 1 and server 2 and server 2 has high configuration.With this algorithm Load balancer will allocate more requests to server 2 to utilize it better. Like Least Connection this also allocates request to the server withthe least number of active connections but the higher configuration server will handle more requests based on the right proportion defined during the Load Balancer setup. Weight proportion is calculated by the server capacity. Server 1 : Capacity X, Server 2 : Capacity : 10X the weight for each server would be 1:10 .

There might be the scenario if you don't want to overload any of the servers for some reason, you may use this algorithm to give extra weight to other servers.

Round Robin

Round Robin is a vastly used and simple algorithm. Load Balancer distributes requests in the cyclic order irrespective of server inefficiencies.

Both the Servers are ready to take request, suppose request comes and load-balancer routes to Server 1 then if a second request comes it will be routed to Server 2. The third and fourth will be routed to Server 1 and 2 respectively in a cyclic order. Even if one of the server has stronger configuration i.e. RAM, CPU etc. As per Round Robin algorithm Load-Balancer will follow the cyclic order.

Algo 4 Weighted Round Robin

Like Round Robin this is also cyclic but the higher configuration server will handle more requests.Rest it is same as Weighted Least connection algorithm i.e. weight would be defined during Load Balancer setup and high weight server will handle more requests.

i.e. if weight proportion to server 1 and server to is 1:10. first 10 requests will go to server and 11th request will go to server 1 and next 10 request will go to server 2 and so on.

Algo 5 IP Hash

This algorithm generates a hash key using client and server IP addresses which means this key would be assigned to client for subsequent requests which assure that the client is routed to the same server that it was using earlier.This algorithm can be used to achieve Session Persistence or Sticky Session.

Session Persistence

In the multiple server environment a user might experience losing cart items during navigation. Persistence session or sticky session is the culprit. As you know Http is a stateless protocol which means in subsequent requests it doesn't maintain any information about the user so to identify server uses client's ip or cookie to track users session. Sticky session is to make sure all the requests goes to the same server which has its user's session tracking information.

Layer 4 & Layer 7 Load Balancing

In Layer 4 Load Balancing , Load balancer decides the server on which it will redirect the requests on the basis of the IP addresses of the origin and destination servers (Network Layer : Layer 3) and the TCP port number of the applications (Transport Layer : Layer 4).

On the other hand Layer 7 Load Balancing decides routing on the basis of OSI Layers 5, 6, and 7 which together makes Http.

To get an overview of OSI Model, please read my post .