To control the distribution of connections and traffic across pool members, Big IP F5 employs a number of load balancing techniques.

There are two main categories into which all load balancing techniques fall:

Ø  Static Load Balancing

Ø  Dynamic Load Balancing

Also we will talk later on about failure mechanisms.

Now we will discuss all the load balancing methods one by one:-

Ø  Static Load Balancing:-

Without taking end-to-end server performance into account, this method distributes connections and traffic according to a pre-defined pattern. In this approach, BIG-IP F5 evaluates the data from the health monitors to determine the status of the server. If any node or pool member is identified as offline, F5 will not send traffic to them.

There are two types of static load balancing methods available which are as follows:-

ü  Round Robin

ü  Ratio or Weighted Round Robin

 

ü  Round Robin:-

The purpose of this strategy or method is to ensure that the load is distributed evenly throughout all of the pool members. In addition to that, it is the default approach for load balancing. The only time that this method is appropriate is when we have servers that host apps that are of equal performance.

ü  Ratio or Weighted Round Robin

By using this technique or method, we are able to adjust the ratio weights such that they correspond to the capacity of the actual servers. In accordance with the ratio weights that have been established on the server, the traffic is load balanced.

Consider the following scenario: we have three servers, and the ratios on each server are assigned in accordance with the hardware capabilities of the servers. Assuming that the first server is given a ratio of three, then the first three connections will be directed to that server, and so on.

When our servers do not have capabilities that are comparable to one another, this type of strategy is typically utilized.

Let's say that we have a pool in which we have two servers with high performance and two servers with low performance. As a result, we have specified the ratio as follows: 3:3:1:1. As of right now, Server 1 and Server 2 will each receive three connections, while Servers 3 and 4 will each receive one connection in accordance with the round robin method.

In the event that we do not define any ratio, it will be 1 by default.

Ø  Dynamic Load Balancing:-

The Dynamic Load Balancing approach distributes the load to each member of the pool based on the performance of the servers or with some other parameters. When utilizing this particular kind of load balancing, the total load that is distributed across the servers can vary during specific time periods.

Below are the methods for dynamic load balancing:-

ü  Least Connections

ü  Fastest

ü  Least Sessions

ü  Ratio Sessions

ü  Ratio Least Connection

ü  Weighted Least Connection

ü  Observed

ü  Predictive

ü  Dynamic Ratio

 

ü  Least Connections

While using this strategy, F5 will distribute the connection between end-Servers and F5 Devices based on the current active connection count. Under the assumption that every server will have the same amount of connections with F5, the connections will be dispersed in a manner that is round-robin. It is possible that the least number of connections is the most relevant in situations when the client connection length fluctuates greatly and round robin could result in uneven loads.

Please note that only active connections will be counted when using this method, connections that are not currently being used will not be counted.

ü  Fastest

 When deciding where to transmit the next client request, the fastest load balancing approach uses the outstanding layer7 request. Using response time to make a decision could seem like the quickest load balancing option, but what exactly is the optimal response time? The speed with which a web server listening on port 80 responds is not factored into the response time to ping. The speed with which the backend database server can populate the web page content is not considered in the syn-ack response to a SYN for a port. In both cases, the fastest load balancing approach does not rely on reaction time but on outstanding layer7 requests.

If servers 1, 2, and 3 all have the same number of unfinished layer7 requests, BIG-IP will round robin between these servers. Keep in mind that server 4 can't be used until all of its pending layer 7 requests reach a similar volume to that of other server.

ü  Least Sessions

In this approach, the connection is distributed by F5 according to the number of persistence records that are kept in the persistence table. Following the transmission of a new request to F5, F5 will examine the Persistence table. The server that has the lowest persistence record will be the one to receive the traffic that was transmitted by F5. 

In the event that the Virtual server is set up to make use of cookie persistence, the F5 functionality will automatically revert to the round robin load balancing mechanism.

ü  Ratio Sessions

The concept of ratio, as well as the least session method, are both utilized in this method.  In order to comprehend this, let us look at the figure below.

The information presented here reveals that server 1 has 10 persistence records, server 2 has 20 persistence records, and server 3 has 25 persistence records.

Server 2 will receive twice as many connections as server 1, and server 3 will receive three times as many connections as server 1. This is the current situation.

As of right now, both Server 1 and Server 2 have already reached the required number of connections in order to completely satisfy the ratio.  On the other hand, server 3 does not have three times as many connections as server 1, which means that any request that is received by F5 will be automatically transferred to server 3.

This will continue until the relationship between the ratio and the persistence record of Server 3 is equal. This means that Server 3 will continue to receive the traffic until the current active persistence record reaches the value of 30. Server 1 will once again be able to receive a new connection once it is finished.

ü  Ratio Least Connection