Victor Yodaiken


    Every computer server has a clock on it and a way to ask the operating system the current time, but  at HFT speeds making sure that time is accurate is technically difficult and critical to business. A trading application trying to figure out the simplest information, such as how long a trading venue is taking to confirm trades, is helpless without good time synchronization. The timestamps on confirmations reflect the clock in the exchange or other venue. Are those ahead of or behind the local clock, or do they wander around? Is the local time accurate? Does it jump forward or even jump backward? Everything becomes even more complicated when multiple servers are involved in a trade, so a robust time synchronization infrastructure is crucial for high frequency traders.

    Time synchronization is hard because when microseconds matter all the imperfections of standard enterprise software, computing platforms, and network technology are magnified. For example, the clocks on servers often run either too fast or too slow or alternate between too fast and too slow. To keep a collection of computers synchronized, a “reference source” can send update messages that tell the computers running trading software how to correct their clocks. However, the update message takes some time to cross the network and this time can vary. This is referred to as Packet Delay Variation or PDV.

    If the trading application runs on server in New York City, the confirmation comes to second server in New York, and the trade itself passes through a collocation server to check against limits before going out to the exchange, and if the clocks on the three computers are not synchronized, it is impossible to derive useful information from the timestamps on trading messages.  To pick just one example, how can you show that your system was not front-running in the absence of synchronized time and reliably time-stamped records?

    The standard “client” software and many of the hardware devices that interpret time updates do not smoothly adjust for corrections or offer the sophisticated filtering needed to compensate for PDV. The easy path is to simply reset the time sources to what the update says, perhaps with a fixed offset to compensate for delay. This means time can jump forward when the update says it is later than the local clock thinks it actually is and even jump backward when the update says it’s earlier than the local clock says. For instance, if the update message gets delayed unusually long because of network congestion it may carry a “correct time” that is earlier than the time registered on the local clock. Imagine what that kind of “correction” can do to a high frequency trading algorithm.

    To further understand why time synchronization technology is so important in high frequency trading, it is important to fully understand some of the difficulties and pitfalls that HFT participants currently face regarding timing in the markets. The most common source of reference time is via a device that gets the time from global positioning satellites (GPS) and sends messages with time updates out to the application servers. There are two standard network protocols for those updates – the older Network Time Protocol (NTP) and the newer IEEE 1588 Precision Time Protocol (PTP). Contrary to some marketing claims both are able to produce a high level of accuracy. With either network protocol, one of the common issues that may arise in a high frequency trading environment is that the network equipment can drop parts of the time protocol or damage the time packets, throwing off system. In another scenario is a network switch or router might fail or introduce timing problems and reference time source can fail silently. 

    Human error is an important factor to consider as well. Sometimes, security engineers who do not know all of the time synchronization packet protocols can inadvertently block part of the protocol when a network switch is being reset or updated.

    In high frequency trading, the difference between success and failure is measured in mere microseconds or even nanoseconds. With that in mind, time synchronization technology has become one of the most operation critical technologies in the high frequency trading sector, especially with market risk levels heightening and an overall increase in regulatory scrutiny. Very simply, time synchronization infrastructure must be a top consideration for every high frequency trader as 2014 approaches.


    The opinions and writing contained in this article are of the author alone and do not necessarily represent those of

    Related content

    News: Arista Networks Unveils Versatile Software Defined Switching Platform for Demanding Applications
    20 September 2012 – Arista Networks
    New Arista 7150 Series Offers 40GbE, Lower Latency,   Flexible Forwarding and VXLAN support  SANTA CLARA, Calif., Sep.19, 2012 – Arista Networks today…

    News: Algo-Logic Systems 3rd Generation TCP Endpoint Achieves Ultra-low-latency of 76-nanoseconds on Stratix V FPGA
    26 July 2013 – Algo-Logic
    U.S.-exchange-certified TCP Endpoints successfully deployed for Accelerated Finance Applications    Santa Clara, California, July 26, 2013 – Algo-Log…

    News: Perseus Launches Precision Time Service in CME Group’s Aurora Data Center
    3 October 2013 – Perseus Telecom
    CHICAGO and NEW YORK, October 3, 2013 – Perseus Telecom and CME Group, the world’s leading and most diverse derivatives marketplace, today announced a new ser…

    Leave A Reply