The Language of Latency
The Language of Latency
By Fergal Toomey, Chief Scientist and Co-Founder at Corvil
“Latency - a measure of time delay experienced in a system, the precise definition of which depends on the system and the time being measured.” Wikipedia.org
All professionals working in financial trading today have heard of ‘latency’ and of its significance in the electronic marketplace. But as the above quote from Wikipedia makes clear, ‘latency’ itself is a technical term whose meaning depends on context.
An exposition of latency-related matters can easily become a thicket of technical jargon to the uninitiated, who also need to be aware that similar terms are often used to refer to subtly different things. In an effort to avoid these pitfalls on LatencyStats.com, we’ve created the following glossary which we hope clearly explains some of the terms we use on the site.
Latency– LatencyStats.com monitors market data latency across NYSE Technologies’ Secure Financial Transaction Infrastructure® (SFTI®) network. This is the time taken by market data traffic to travel from the feed assembly points where the feeds are created, to the point in the access center where the data is handed over to trading clients. The latency values are presented in units of microseconds (µs) – millionths of a second.
Latency Transparency– for an exchange or a service provider, latency transparency means sharing information about latency performance with your trading clients. The goal is to better inform clients about the nature and quality of the products and services they use.
Millisecond (ms) - One millisecond is 1 one-thousandth of a second – a very short timescale. It takes five whole milliseconds for a flying honey bee to complete one full up-and-down wingstroke.
Microsecond (µs) - An even shorter timescale, one microsecond is 1 one-millionth of a second (or 1 one-thousandth of a millisecond). Think of a friend standing at one end of the Brooklyn Bridge in New York while you stand at the other. If she waves, it will take just over 6 microseconds for the light reflected from her arm to reach your eyes. Latency values on LatencyStats.com are all presented in units of microseconds.
Average, Peak and 99.9% Latency -The delays experienced by market data messages change all the time, primarily because of changing activity patterns in the feeds themselves. Because of this, there is no single “latency number” that describes the performance of a feed. The actual latency values are spread over a range, which we describe on LatencyStats.com in terms of the average and peak (maximum) values seen during the selected timeframe. We also present the 99.9th percentile latency. This is the latency value that is exceeded by only 1 in every 1000 market data packets.
Volume -The volume statistics presented on LatencyStats.com measure the quantity of market data being delivered in the feeds. Feed volume is significant for us because it can affect latency – if the volume of the feed exceeds the capacity of any processing or forwarding systems along its path, then a backlog will build up which adds latency. For users wishing to receive the feed data as quickly as possible, even a temporary backlog lasting a fraction of a second is a cause for concern. For this reason LatencyStats.com provides information about peak volume rates over a range of timescales down to one millisecond duration.
Average, Peak 1-Sec and Peak 1-ms Rate -The “Average Rate” represents the average volume rate over the whole of the selected timeframe, measured in megabits-per-second. The “Peak 1-Sec Rate” is the highest rate that was seen during any 1-second period during the timeframe, and the “Peak 1-ms Rate” is the highest volume rate seen over any 1-millisecond period.
Bandwidth -The term ‘Bandwidth’ refers to the amount of data that can pass through a given point in the market data path per unit time, measured in units such as kilobits-per-second. Bandwidth is not quite the same thing as speed, although the terms are often used synonymously. Nevertheless, a lack of sufficient bandwidth at a bottleneck point can slow the data down, by forcing it to queue when conditions are busy. On LatencyStats.com, the bandwidth figure presented for each feed is the minimum access connection bandwidth needed to prevent queuing delays from ever exceeding one millisecond.
Sequence Gaps - While speed of delivery is important for market data, making sure that all of the messages are successfully delivered in the correct order is clearly just as important. Missing messages, or messages that are out-of-order, can be detected using the “sequence numbers” embedded in each market data message. A sequence number “gap” indicates that some messages may have been lost on their way to the end consumer. When this happens, consumers typically must ask for the messages to be re-sent in a process that adds considerable latency to the final delivery time.
Gap Rate - LatencyStats.com provides a measurement of the gap rate seen in each market data feed at the point where it is handed over to clients. The gap rate is calculated by decoding the sequence numbers in the delivered messages to determine the percentage that are missing during the selected timeframe.