Our announcement last week was an exciting milestone for those of us who started at 29West supporting the early high-frequency traders from 2004 to 2006. Last week, we announced the next step in a 10 year effort that has now seen us set the bar for low latency messaging lower by six orders of magnitude in Version 6.1 of Informatica Ultra Messaging with Shared Memory Acceleration (SMX). The really cool thing is that we have helped early customers like Intercontinental Exchange and Credit Suisse take advantage of the reductions from 2.5 million nanoseconds (ns) of latency to now as low as 37 ns on commodity hardware and networks without having to switch products or do major rewrites of their code.
But as I said in the title, what does it matter? Does being able to send messages to multiple receivers within a single box trading system or order matching engine in 90 ns as opposed to one microsecond really make a difference?
Well, according to a recent article by Scott Appleby on the TabbFORUM, “The Death of Alpha on Wall Street”* the only way for investment banks to find alpha or excess returns is “to find valuation correlations among markets to extract microstructure alpha”. He states “Getco, Tradebot and Renaissance use technology to find valuation correlations among markets to extract microstructure alpha; this still works, but requires significant capital.” What that extra hundreds of nanoseconds that SMX frees up allows a company to do is to make their matching algorithms or order routers that much smarter by doing dozens of additional complex calculations before the computer makes a decision. Furthermore, by allowing busy software developers to let the messaging layer takeover integrating software components that may be less critical to producing alpha (but very important for operational risk control like guaranteeing that messages can be captured off the single box trading system for compliance and disaster recovery) they can focus on changes in the microstructure of the markets.
The key SMX innovation is another “less is more” style engineering feat from our team. Basically SMX eliminates any copying of messages from the message delivery path. And of course if the processes in your trading system happened to be running within the same CPU on the same or different cores, this means messages are being sent within the memory cache of the core or CPU. The other reason this matters is that because this product uniquely (as far as I know) allows zero copy shared memory communication between Java, C, and Microsoft .Net applications, developers can fully leverage the best features and the knowledge of their teams to deploy complex high-performance applications. For example, this allows third-party feed handlers built in C to communicate at extremely low latencies with algo engines written in Java.
So congrats to the UM development team for achieving this important milestone and “thanks” to our customers for continuing to push us to provide you with that “lagniappe” of extra time that can make all the difference in the success of your trading strategies and your businesses.
Right before Christmas, I was delighted to read about the proposed merger between the New York Stock Exchange and Intercontinental Exchange. ICE and NYSE have been customers that we on the Informatica Ultra Messaging team have been working with for several years. NYSE Technologies leveraged our high performance messaging as part of their direct feeds market data solution that lowered latencies across dozens of Wall Street firms around the globe. (more…)