Comparing load balancers is simple, but not easy

Throughput is the amount of data, usually referred to as traffic, being (or capable of being) transmitted and received ‘over the network’ by a device such as a load balancer or server, per second. You may also be able to monitor the throughput of a VIP but this will always be limited by the capabilities of the device it exists on.

Typically the maximum throughput of a load balancer is determined by the maximum throughput (or speed) of all of its network interfaces combined. Our Enterprise Prime hardware appliance has four interfaces with a speed of 1Gbps so its maximum possible throughput is 4Gb.

However, you should also be aware that many load balancer vendors place artificial license restrictions on their devices, meaning that although the network interfaces are capable of higher throughput the appliance is still throttled for commercial reasons.

As an additional note, protocol overheads mean you’re only likely to actually achieve around 80% of the maximum theoretical throughput of an interface. It’s also worth keeping in mind that cloud based host virtual network interface throughput may be limited by factors other than the reported speed of an interface. So while the operating system may be presented with a virtual 1Gb network interface, its actual maximum possible throughput may be far lower.

SSL/TLS Transactions Per Second. Establishing SSL/TLS security for a new connection is a CPU intensive operation and the number of new connections that can be handled at any one time is limited by the power of a device’s CPU(s). The key size used has a significant impact (using contemporary ECC keys, which are much smaller, will result in the highest TPS). Once a connection is secured and established things become far less taxing. A device can support hundreds of thousands or even millions of active SSL/TLS secured connections, but its capacity to establish new ones is limited by this figure.

HTTP RPS is number of incoming HTTP Requests Per Second that can be processed by a load balancer or server. As small as they usually are in comparison to responses, requests contain a lot of information that must be parsed to ensure they are valid, do not present a security threat and to make load balancing decisions based upon any number of components within the request. For this reason, the maximum number of RPS is significantly lower than the maximum number of connections a load balancer supports.

A connection is, typically, an established communication channel between a client (a user using a web browser for instance) and a server or site, although in reality this is almost always a load balancer. Where web based traffic is concerned, a client browser will typically open multiple, parallel connections to the same site (Chrome will open up to six for instance) in order to improve performance — one user rarely equals one connection.

Concurrent connections: this is the number of established connections between one or more users and a single server or load balancer. As we know the most common browser could open up to six connections to a site.