Overview
On the one hand, we are moving towards virtualization and dividing physical servers into virtual ones, on the other hand, sometimes we may lack the performance of one server and have to combine them into a cluster. Multiple servers running the same service can improve not only overall performance, but also reliability. Load balancers are used to ensure that a user application can work with a group of servers as a single server. They look like a single server to the user application, to which it connects, but in fact they only transmit messages from the client to the "real server" and back. This technology is similar to balancing the speed of Internet access, which I described earlier, but distributes data streams not to the Internet, but to the internal servers of our organization.
Let's take a closer look at the picture above. We have users. They can connect from the local area network or from the Internet. We need to provide an email service for them. There are so many users that one server cannot cope with the flow of their requests. In this regard, a cluster of servers is configured, each of which can serve each client. In order for the client not to have to choose the "real server" himself, all the servers in the cluster use one common address to which the clients connect. These connections are shown with black solid arrows. The firewall acts as the owner of the public address. He also creates a virtual mail server inside himself with this public address.
When a client connects to a virtual mail server, the firewall picks one of the "real servers" and forwards the connection to it. These connections are shown with black dashed lines. As a result, each client connects to some "real server", moreover, the firewall chooses with which one (there are different algorithms for this). The real client connection is shown with green dashed lines.
As a result, the client connects to the firewall, thinking that this is one single mail server, and the firewall serves this connection in such a way that the server from the cluster can act as this server. For the client, the address always remains the same, the real server address is masked. For the server, the firewall acts as a router when communicating with the client, so everything also works as it should. There remains only a question of interaction of servers in a cluster with each other. If they use a common set of information (for example, letters sent to the client), then it is necessary to ensure synchronization of this information between them (this is no longer the task of the firewall).
