Comparing speed tests

Tests that vary by speed can be compared to clearly show how throughput, latency, and send queue size change as the rate limit changes. This can be used to quickly identify the exact bandwidth provided by a leg—that is, the speed at which the modem or ISP router begins to queue packets. Avoiding queues helps maintain quality of service and performance of the IDMPQ packet distribution algorithm.

To compare a set of tests, click the check box beside each test you want to compare, then click “Compare tests”.

image0

You can quickly run multiple tests at different speeds by using a value such as “6.0,6.1,6.2,6.3” in the rate limit field when starting a test.

You are then presented with a single chart showing the average throughput, latency, and send queue values for each of the tests. The X-axis represents the rate limit for each test.

image1

As the rate limit (the value on the X-axis) increases,

  • Throughput should increase until the rate limit reaches the speed provided by the leg. At that point, it should remain constant.
  • Latency should remain low until the rate limit reaches the speed where traffic is queued by the modem or ISP router. Then it should increase sharply.
  • Send queue size should remain high until the rate limit reaches the speed where the ISP queues traffic. Then it should decrease to zero.

To find the speed at which the ISP begins to queue traffic, identify the rate limit where throughput plateaus, latency increases, and send queue size decreases. If each indicator changes at a slightly different rate limit, select the limit just before latency increases—the highest limit where latency is still low.

In the example above, the highest rate limit that still offers relatively low latency is 6.1 Mbps.

Limitations

  • There is little point comparing tests with different targets (i.e. tests for two different legs in a bond, or a leg test and a bond test), protocols, or directions. There are too many differences between those types of tests to compare them meaningfully.
  • You can compare multiple tests with the same rate limit, but this makes the graph harder to interpret.
  • Tests with no rate limit do not appear in the chart.