Exegy Market Data Blog

The Language of Low Latency- why speed isn’t good enough

Dave Mishoe from SunGard wrote a fine article in the TABB Forum entitled Low Latency: What’s Your Hurry? http://tabbforum.com/opinions/low-latency-what%27s-your-hurry.  The essay triggered a number of thoughts on the same subject. They are slightly random, so please bear with me.

Mishoe started with the observation that there was a moment in the past where he heard the term “low latency” for the first time. He couldn’t remember when that was, but he did know that it was an important moment. For my part, having been involved in market data for over 20 years, I have exactly the same recollection. In other words, I can’t remember when “it” happened. And shame-facedly, I can’t even remember the year. However, I can probably work out the tipping point for the industry by deduction.

My guess is that the axiomatic time, the momento magnifico, almost certainly occurred in 2005 when Reg NMS was introduced.  

Reg NMS sanctified the current market structure of competing liquidity pools and acted as a starting pistol for the ECNs to develop enough liquidity to become full fledged exchanges.  Reg NMS also marked the take-off point for an extraordinary growth in market data from the stock markets as they organized their banks of computers into ever faster trade matching engines, with ever larger books of orders and constantly improving trade execution times. Everything became faster. Low latency ruled!

How low do you go?

Five years later, low latency is a widely used term. But it has to be said that the phrase has been much abused.  In other words, when is fast computing low latency computing?  When does low latency become ultra low latency?  How do you measure speed?  From where to where exactly?

In 2010, we’re at a new tipping point where the phrase is getting some better definition, and some firms and have been created simply to provide reliable statistics on disparate computer systems. More on that later.

Low latency and cheating

There is a dangerous current of thinking that condemns low latency trading, or God forbid, ultra low latency, as a kind of cheating.  But as Mishoe says, the entire history of communications from the telegraph, to the telephone, fax and email have all been about lowering latency.  That whole debate is not worth addressing here, as so much ink has already been spilt.  But I have a theory that the very term “low latency” has amplified the counter reaction among less agile firms who cannot abide the fact that those who have invested in technology have won a strategic advantage.  It is much more acceptable to complain to politicians and the general public about low latency advantages than speed advantages.  “Low latency” just sounds suspicious to the untrained ear.

Mishoe also raises a great point that checking latency means different things to different people and firms. Some are focused on dealing with outlier trades and quotes, others with the quest for the fastest or some combination thereof. There are several efforts in progress right now that attempt to deal with the prickly problem of measuring latency and providing standards so that we don’t all have to reinvent the wheel. Here’s a few that spring to mind.

STAC Research is one approach where firms can subscribe to special reports on technology performance http://www.stacresearch.com/.  Exegy used STAC way back in 2007 to cross check our results on the second generation Exegy Ticker Plant. I think that Exegy was the first market data system independently tested with microsecond performance. (I am happy to be corrected on this)

There is a small squadron of boutique firms specializing in the difficult science of measuring speed for trading systems at incredibly fine increments. Our friends at Corvil are a prime example http://www.corvil.com/.

There are two well informed discussions in the industry. These working parties are focused on establishing industry norms so that we can compare latency stories. FIX Protocol Limited has a working group on the subject http://www.fixprotocol.org/working_groups/fipl led by Henry Young of TS-Associates and Nicolas Morrison from Nomura.  The Financial Information Forum’s Capacity Working Group started to talk about latency measurements in October. The conversation will continue in future capacity meetings. See http://www.fif.com/md_capacity/ for further details.

In conclusion, it is probably too late to revert the general conversation to questions of mere speed.  Speed just isn’t fast enough. Low latency conveys the concept of urgency and science. The use of the term has been abused by marketers and we’re stuck with low latency for the foreseeable future. However, we will shortly see much improvement in the understanding of what it means by placing measurement standards on the table. Then we’ll really know what fast means. Unfortunately, it won’t stop the accusations that using speed is cheating. For that, I just don’t think there is a cure.

Happy Holidays.