Low-Latency Trading Infrastructure 101:

Measuring Trading Latency

Measuring Wire-to-Wire and Tick-to-Trade

In a trading system, everything unmeasured between input and output hurts performance.

Trading latency is measured wire-to-wire and tick-to-trade:

• Wire-to-wire is tapping the network cables outside the system to record timestamps. This incorporates everything in the system that contributes to latency, including network hardware.
• Tick-to-trade is when the timestamps are taken. It is measured from the first byte of the market data packet seen on the network cable (tick) to the first byte of the resulting order packet going out (trade).

Record timestamps with a Layer 1 switch…

The taps and potentially timestamping can be done by a Layer 1 trading switch such as the 7130 Connect Series produced by Arista Networks (formerly Metamako) or the Nexus 3550-F Modular Layer 1 Matrix Switch made by Cisco (formerly Exablaze.)

Layer 1 trading switches connect an input port to one or more output ports with very low latency. Data typically takes single-digit nanoseconds to travel from the switch’s input port to an output port. An input can be connected to many outputs, for example, to simultaneously distribute latency-sensitive market data to multiple co-located servers.

A Layer 1 switch can be treated as a patch panel, connecting the input of one port to one or more output ports. It can also tap, record, and timestamp network data flow.

… or use a server to record network traffic

Once the Layer 1 switch has tapped the data, an alternative is to send the traffic to a server for an accurate timestamp and capture, for instance, by using Endace hardware to timestamp and save the data to disk with nanosecond precision.

It is crucial to use these measuring techniques from the outset of development to monitor the impact of design changes when building for the lowest possible latency.