T-O-F Latency Measurement

Measure latencies with 31 picosecond accuracy: detect the difference of 7 mm of fiber

There are two distinct tasks in current day networking that use timestamps: traffic timestamping and network efficiency measurement. The first is about collecting dynamic data for future storage and analysis for which timestamping is the natural solution. The second is aimed at gathering maintenance information on the state of the infrastructure, such as hardware processing delays and infrastructure induced latencies. In this case, timestamping was a historical choice for lack of a better option.

The standard setup for measuring infrastructure latency (e.g., top-of-the-rack switch with its cabling) includes a traffic generator, a fiber tap to split the traffic from the generator, and a server with a network card that has two ports and is capable of timestamping the traffic on both. The traffic is split in two: one line goes straight to the server, the second goes through the DUT. Then each connects to one port of the timestamping NIC. The difference between two timestamps of the same packet is considered as latency added by the DUT.

There are significant downsides to this approach. The first is economical: the hardware that is used for the measurement is expensive. The second is accuracy: the measurement precision is determined by NIC timestamping precision. Currently, the highest precision is around 250 picoseconds. This is acceptable for traffic timestamping, but as measurement accuracy, it allows a drift of about 76mm (3") of copper cable.

Is there a better method?

A technology patented by LDA uses a radar approach for measuring latency in a network. The radar method is simple: send the signal into a medium that will loop it back, receive the reflected signal, and measure its “Time Of Flight.”

LDA's Time-Of-Flight latency measurement system highlights:

Products

LDA NeoTap