Latency, wire-to-wire

As the term "wire-to-wire" suggests, it measures the duration from when a message is received from the network (taken off the wire) to when a response message is sent back to the network (put back on the wire). Wire-to-wire latency is often reported in various percentiles, such as the 0th, 50th, 90th, 99th, and 100th percentiles. This approach highlights the importance of tail latency, which can be as significant, if not more so, than average latency figures. Vendors typically report at least the median or mean wire-to-wire latency for marketing purposes.

Different assumptions and considerations can affect how wire-to-wire latency is measured, and not all vendors apply these consistently. Key considerations include:

• Timestamps: Whether the timestamps are recorded at the start of frame (SOF) or end of frame (EOF).
• Throughput: The rate at which data messages are transmitted.
• Message size: The size of the payloads being sent.
• Application logic: The extent of application or business logic processing performed by the trading system in between message handling.
• Protocol decoding and encoding: If the trading system needs to decode market data and encode an order message, the choice of trading venue protocols can significantly affect performance. For instance, FIX-based protocols generally incur more overhead compared to proprietary binary protocols.

Industry standards like the STAC-T1 benchmark are widely recognized for measuring wire-to-wire latency, providing a consistent framework for comparison across systems.

When it's not specified, wire-to-wire latency is often used assumed to be tick-to-order latency, which is a more specific of wire-to-wire latency where the inbound message is a data message and the outbound message is an order message. However, it should be noted that the term "wire-to-wire latency" may be used with more generality: for example, a trading strategy may react not only to market data messages, but also order acknowledgments as well, so the unconditional distribution of the strategy's wire-to-wire latency will include latency samples on both order-to-order and tick-to-order events.

Hardware timestamping cards from vendors like Napatech, capture appliances from companies such as Endace and Corvil, and switches with timestamping capabilities like the Arista 7130 series are commonly used to measure wire-to-wire latency. However, due to the cost and complexity of setting up precise hardware measurements, some practitioners may measure to socket-to-socket latency instead, which only requires a single server and can be done more easily within software.