The Gateway to Algorithmic and Automated Trading

Promoting transparency through data analysis on latency

Published in Automated Trader Magazine Issue 31 Q4 2013

Technology is one of the most important cornerstones of electronic markets and, more broadly, of today's global markets. As technology allows for increasingly complex systems from a design standpoint, it is essential to get a clear view on each step inside a system to understand outcomes. Gathering data at discrete points within such systems is a practical way to break down the complexity and deliver digestible information.

By Markus Löw, Head of Monitoring, Vassilis Vergotis, EVP & Head of Americas at Eurex Exchange and Fabian Rijlaarsdam, Monitoring Analyst
By Markus Löw, Head of Monitoring, Vassilis Vergotis, EVP & Head of Americas at Eurex Exchange and Fabian Rijlaarsdam, Monitoring Analyst

Technology is one of the most important cornerstones of electronic markets and, more broadly, of today's global markets. As technology allows for increasingly complex systems from a design standpoint, it is essential to get a clear view on each step inside a system to understand outcomes. Gathering data at discrete points within such systems is a practical way to break down the complexity and deliver digestible information. Within the area of financial market technology, especially in the case of exchanges and alternative trading platforms, transparency is essential for monitoring the fairness and market structure of financial markets:

• Data analysis on latency is important in this matter, as it provides insight into the technical infrastructure.

• Latency statistics are essential to provide users, operators and supervisory authorities detailed information on the performance and stability of technical infrastructure.

Despite the common perception that these statistics are valuable and important in understanding developments in market structure, two problems exist:

• Firstly, robust statistics are not broadly available yet.

• Secondly, and perhaps more important, is that many people do not fully understand the data that is available. The reason for this is the deep dependency on in-depth knowledge of the technical infrastructure that is used.

This article aims to increase understanding of latency statistics by expressing our view on latency and walking readers through the type of latency monitoring capabilities that Eurex Exchange offers to its direct users. We also include a detailed overview of the large number of measurement points that we have embedded into our T7 system to allow us to continually enhance our technical infrastructure and enable our Trading Participants to help us in doing so.

Furthermore, our latency monitoring tools are helping our Participants to optimize and adapt their trading strategies to the market and infrastructure available. Eurex Exchange's latency, transparency as well as the detailed monitoring capabilities have been well-received by Trading Participants across the board and have helped the Exchange and its Participants to collaboratively solve problems. Most recently, these figures have proven that our new T7 trading architecture has drastically reduced latency and increased consistency in our technology.

Achieving a common definition of latency

As a starting point, we refer to latency as the time between the arrival of a message on our application space 'gateway' and the send-time of the respective message-response, which is commonly known as the (exchange-) round-trip time.

Graph 1 represents the distribution of these exchange-round-trip times on Eurex Exchange's T7. The graph depicts transaction latencies in microseconds. The probability density function of round-trip latencies is shown on the left Y-axis and the cumulated probability is shown on the right Y-axis. When analyzing the graph, one can look at a number of variables: the mean, median, standard deviation or tail (i.e. higher percentiles of the distribution).

When looking at data, the average is an easy figure to grasp - and therefore often used. However, it can be rendered imprecise if the measurement cited is an average collected from a number of different products and/or time frames. Therefore, in the remainder of this article, we will mainly look at the tail of the latency distribution shown in graph 1, as these are the cases in which Participants run the highest risks due to being uninformed of the status of their transactions.

There are several plausible explanations with regards to the occurrence of such a tail:

• On the one hand, a technical issue delays transactions coming in, causing Participants to unnecessarily run higher risks.

• On the other hand, if a certain market situation, e.g. a scheduled news event, causes Participants to behave in a correlated manner, the delays still imply extra risk for Participants, but this risk is introduced by the physical limits of the exchange system: the better the system, the less risk Participants run.

Either way, latency figures or other data analysis on transaction traffic help both Participants and us to understand the robustness of latency in the system. From the graph we can see that, on an average day, over 90% of the transactions in Eurex Exchange's T7 have a round-trip time latency of less than 400 microseconds.

Graph 1: Roundtrip latency distribution of Eurex Exchange’s T7
Graph 1: Roundtrip latency distribution of Eurex Exchange's T7

Gauging T7

Since the introduction of Eurex Exchange's T7, latency statics have demonstrated that T7 is significantly faster than the Exchange's previous trading system. Average (request-to-response) round-trip latency has been reduced from approximately 3 milliseconds on the previous trading system, to approximately 250 microseconds on Eurex Exchange's T7.

Graph 2: Average request-response round-trip times, Eurex Exchange’s T7 versus its predecessor
Graph 2: Average request-response round-trip times, Eurex Exchange's T7 versus its predecessor
Graph 3: Median and higher percentile request-response round-trip times, Eurex Exchange’s T7 versus its predecessor
Graph 3: Median and higher percentile request-response round-trip times, Eurex Exchange's T7 versus its predecessor

However, faster average round-trip times are certainly not the only achievement introduced by Eurex Exchange's T7. The new trading system has also increased their consistency. Many of our Participants have cited the importance of consistency; to some extent, they assign an even higher priority to it than to speed. Consistent round-trip times help participants to better implement their strategies, as this consistency mean less operational risk. For this reason, we now take a look at the round-trip distributions for Eurex Exchange's T7 versus its predecessor to demonstrate our achievements in increasing consistency. The round-trip latency spectrum of Eurex Exchange's T7 is much tighter than that of the previous trading system. Graph 3 above shows that the ratio between the 95th percentile and the 50th percentile has been reduced from a ratio of 15 to a ratio of 2.5 and the 95th percentile of all round-trip latencies on Eurex Exchange's T7 is now even below the median (50th percentile) of all round-trip times of the previous system.

According to a number of sources, including Automated Trader's extensive 2013 Global Trading Trends Survey Report, many Participants in the market have taken the decision to withdraw from the "latency arms race", judging incremental improvements in round-trip times to be just too costly. Against this backdrop, the consistency improvements triggered by Eurex Exchange's T7 are more important to a wide range of market participants that deploy less latency-sensitive investment strategies.

Monitoring tools at Eurex Exchange

Eurex Exchange supports the goal of promoting system transparency in the industry - and we've started in our own backyard. Our wide range of monitoring tools is designed to give users extremely granular and detailed timestamp information. Furthermore, Exchange Participants do not only receive their own data, they also have access to the "best in class" data values for each measuring point in order to benchmark their performance.

Within the round-trip measurement discussed earlier in this article we implemented a series of smaller measurements on numerous points within our hardware and software, in order to provide as many independent measuring points as possible and practical. By doing so, we provide our users access to more data to enhance their own performance analysis. A diagram of the topology and explanation of the measurements taken at each location is provided opposite.

GRAPH 4: Timestamp overview

Explanation of timestamps in Graph 4:

t_1: the Exchange Participant sends a transaction - available to Participants

t_1': the Exchange receives the transaction on its network - unavailable to Participants

t_3: the gateway receives the transaction - available to Participants

t_3': the gateway forwards the transaction to the requested partition - available to Participants
as of 25 November 2013

t_5: the partition receives the transaction; time-priority is assigned - available to Participants

t_7: The transaction is processed by the matching engine - available to Participants

t_6: response is sent towards the gateway - available to Participants

t_4: the response leaves the gateway towards the Participant - available to Participants

t_4': the response is received by the gateway - available to Participants as of 25 November 2013

t_2: Exchange participants receives its response - available to Participants

t_2': the response leaves the Exchange network - not available to Participants

t_8: Market data is sent out via multicast - available to Participants

Graph 4

Transparency promotes trust as well as cooperation

Wolfgang Eholzer, Head of Trading IT, explains that offering latency monitoring tools has benefitted market participants and their introduction has been immensely positive for the Exchange. He says: "When we first introduced our monitoring tools, we quickly learned that empowering our Participants to monitor their own latency, was an immensely productive step. Our Trading Participants' IT staffs proactively contacted us with questions and observations - and some helpful constructive criticism. As a result of our openness, we were able to work together to solve both perceived and actual problems and optimize our respective networks in an extremely efficient manner. This teamwork continues today."

For more information on Eurex Exchange's latency statistics, please contact us via:
hft@eurexchange.com

www.eurexchange.com