Lowering Latency – How low is 'low enough'?


By Bill Hebert, Vijay Kedia, Donal Byrne
At the recent FPL Americas Conference, Bill Hebert, FPL Americas Education and Marketing Committee Co-chair moderated a panel of industry experts on “Latency limbo: How low can you get?.” The panelists’ insightful observations were so well received by the delegates that we decided to bring them to you. Here, two of the panelists, FlexTrade System’s Vijay Kedia and Corvil’s Donal Byrne, communicate their insights in response to Bill’s questions.
Bill Hebert: How do you best define ‘latency’ as it pertains to the electronic trading world and the issues different firms such as yours are facing?
Vijay Kedia (FlexTrade):
In the world of electronic trading, latency is the delay between receiving knowledge of a change in the market and acting upon it. During this time, information travels through both software and hardware, each element along the path introduces a measurable delay before a message reaches its destination. Latency is inevitable. The biggest challenge for vendors of high frequency algorithmic trading platforms is balancing rich functionality with the processing cost incurred because of it. Everything comes at a cost.

Donal Byrne (Corvil):
While the term ‘latency’ has a specific technical definition, it is important to remember that in electronic trading it is used as a proxy for the question “How fast am I trading”? While you might think this is a simple enough question, it is actually quite complex and proving very difficult for traders to get consistent and useful answers. There are three main reasons for this:

– knowing absolute latency is necessary, but not sufficient to determine if you will be successful in high frequency trading. Knowing latency relative to your competition is the key. This is often difficult to achieve.

– today latency is not usefully described in our industry. A single published number is insufficient and often misleading to use in describing the latency performance of an electronic trading infrastructure. In addition, latency should be measured under load conditions that represent intended use. What is needed is the measurement and publication of latency distributions, measured during busy trading periods, e.g. during the busy 1 millisecond of the trading day.

– latency measurement is required on an end-to-end basis for the electronic trading loop. This includes both market data paths and order execution paths. Unfortunately, the measurement of end-to-end latency in a trading loop involves two significant issues: How to measure latency across infrastructure that is owned and controlled by multiple parties, i.e. trader, venue and service provider? How to achieve microsecond accuracy latency measurement across the wide area?

Bill Hebert: What are some latency myths and misperceptions? How are firms using latency management/measurement as a “sales” tool and/or strategy?

Donal Byrne (Corvil):
The single biggest myth and misperception is that ‘Latency’ is a single constant number. We are all familiar with the typical claims:

  • Technology Vendor – “The feed handler is benchmarked at 10us”
  • Market Center – “We can execute an order within 350us”
  • Data Provider – “The distribution latency of our direct feed is 2ms”
  • Telecom Provider – “Our latency is less than 55ms transatlantic”

Advertised latency numbers have taken on major commercial significance in the world of high frequency trading, as many in the industry publish numbers in an attempt to show their service in a favorable light and to demonstrate superior performance over competition. Unfortunately this method of describing latency does little to help end-users understand the true performance of the underlying low-latency service. Market pressures are such that few dare to offer latency information that brings real insight and transparency to latency performance due to the fear of “appearing slower” than the competition. As a result, most informed customers of these services are forced to ignore the published latency claims and look to measure and benchmark latency service levels independently.
Secondly, ‘latency’ is not a constant. It is a lot like the weather, always changing. If you don’t like a particular latency number, then wait a millisecond. It will change. Why does it change?
Latency is not simply dependent on geographical distance alone. It is also dependent on factors like traffic load, network bandwidth, and processing capacity. If any of these factors change, then latency will change. The load on electronic trading systems is highly variable. We know that intraday patterns often show higher volumes at market open and close. We also see large variations in load at micro timescales. These are known as microbursts. Active components that process trading messages like trading engines, line handlers, gateways, firewalls, switches and routers can often be temporarily congested by these sudden microbursts which adds even further to the latency.
Vijay Kedia (FlexTrade):
The reality of electronic trading is that ‘time equates to the opportunity to profit’. As hardware and software gets cheaper and faster, the window of opportunity shrinks, and may be lost to the quicker, more responsive trader. Simply reducing latency will not guarantee profit, since the trading decision is still dependent on trader or strategy, but it does allow for a competitive edge. Additionally, hardware and connectivity matters; no matter how quick and efficient the system promises to be, if the entire path to market is not evaluated, the perceived gains within software will be negated by saturated market data bandwidth or overloaded switches on the way to the exchange. Latency numbers given in marketing materials are often misleading or have implicit assumptions, which do not accurately reflect realistic trading scenarios, and should always be taken with a grain of salt. Without a full understanding of how the metrics were generated, the numbers are just as useful as lottery numbers on the back of a fortune cookie.

Bill Hebert: Which has greater impact on the end client and how do you address the differences between issues with trading (order placement and execution reporting) and market data (quotes, trades, etc.)?
Vijay Kedia (FlexTrade):
Both order/execution management and market data are interconnected, but the weight placed on each is different depending on the trading strategy. While the goal is always to make each component of the system as fast as possible, the software vendor cannot place the emphasis on the client’s behalf. Providing the option to choose which component should be prioritized enables the client to manage the level of latency they are willing to accept. When designing the system, a modular approach allows for the most flexible trading solutions.

Donal Byrne (Corvil):
The key question of “how fast am I trading” depends on two further questions: How fast is market data being delivered? How fast can I execute to a given market center?

Both aspects are equally important in high frequency trading. In our experience, more focus has being placed on optimizing the speed of order execution while latency measurement and optimization of market data has sometimes taken second priority. As a result, we often come across situations where reducing order execution latency does not always produce corresponding improvements in fill rates. This can be due to excessive market data latency relative to competing traders.
Bill Hebert: In the EMS, DMA, optimal liquidity seeking environment we are in today are not the two (trading and market data) even more interdependent and timeliness/data synchronization effectively mandatory?

Donal Byrne (Corvil):
As explained above, the latency of order execution and market data are both critical to improving fill rates and interdependent in determining the effectiveness of a high frequency trading strategy. However accurate measurement of end-to-end market data latency is a bit more challenging than measuring order execution latency. This is due to the one-way, multicast nature of many direct market feeds. Microsecond accurate latency measurement requires precise time management/ time-stamping at both source and destination. This can be achieved using synchronization technology like GPS or PTP in all measuring or time-stamping devices. However, this often proves challenging to provision in practice as it requires coordination and collaboration between multiple parties involved in managing the infrastructure along the end to end market data distribution path. As a result, embedded timestamps in market feeds are often inaccurate, lack the required precision and use of them in computing latency can result in negative numbers. This is one of the critical issues that need to be addressed in high frequency trading infrastructure.

Vijay Kedia (FlexTrade):
When making trading decisions, the most accurate, up-to-date information is of paramount importance. Having extremely fast order entry, but delayed market data is no better than sending market orders at random time intervals. Most trading strategies make an implicit assumption about having zero latency in market data and order entry, so reducing these delays is essential to generating meaningful trades.