This report is used to visualize the how the performance of the broker is changed over time as new code changes are introduced. Click on any data point in the charts to get the data point's exact value and git commit version.

Producers are sending non-persistent messages and do NOT wait for a broker ack before sending the next message. Consumers auto ack.

General

Max Producer Rate

The maximum rate that a single producer can send to a broker. The producer is setup on a topic with no consumers. The broker just drops all messages the producer sends it. Use this to get an idea of what is the maximum transfer rate between two endpoints on your machine.

Queue Performance

This section examines the performance of queues also known as the point to point messaging domain.

Partitioned Scaling

Compares how well the broker scales as partitioned load is increased on it. Each destination has only one producer and one consumer attached using small 20 byte messages. This should scale up with more cores. Keep in mind that the load test is running the load clients and the broker on one machines so about 1/2 the cpu resources are being used by the load clients.

Destination Contention

When there are many consumer or/and producers on one destination, it can become a bottleneck. Shown data points are the total consumer rate for scenario.

Fairness

When you have multiple homogenous consumers or producers, it's ideal if the the broker treats them all fairly and sends or accepts the same number of messages from them. This chart shows the standard deviation of the number of messages produced or consumed. The lower the number the better.

Topic Performance

This section examines the performance of topics also known as the publish/subscribe messaging domain.

Partitioned Scaling

Compares how well the broker scales as partitioned load is increased on it. Each destination has only one producer and one consumer attached using small 20 byte messages. This should scale up on machines with many processing cores. Keep in mind that the load test is running the load clients and the broker on one machines so about 1/2 the cpu resources are being used by the load clients.

Destination Contention

When there are many consumer or/and producers on one destination, it can become a bottleneck. Shown data points are the total consumer rate for scenario. When looking at the numbers, keep in mind that topics replicate/broadcast every messages to every consumer so there is much much higher message load when you have many consumers attached.

Fairness

When you have multiple homogenous consumers or producers, it's ideal if the the broker treats them all fairly and sends or accepts the same number of messages from them. This chart shows the standard deviation of the number of messages produced or consumed. The lower the number the better.