Trading Technology

01:38 PM
Connect Directly
RSS
E-Mail
50%
50%

Did JVM's "Warm-Up" Cause Knight's Collapse?

Some say that Knightesque “flash trades” may result from a known behavior of many Java Virtual Machine solutions and at least one vendor announces a fix.

With the buzz around the recent introduction of the NASDAQ’s “kill switch” tool (PDF) the ghosts of the August 2012 Knight Capital Group trading debacle are again haunting the minds of many on The Street. Although participating firms can count on kill switches like NASDAQ’s to shut down trading based on each participant’s pre-defined limits, this remedy is almost akin to closing the door after the horse has already left the barn. In other words, such switches don’t address the underlying issue, which means firms risk exposure before the switch activates. How much exposure depends upon what limits a firm sets. While there are no public statements by Knight, or other insiders, on the exact nature of the 2012 “technology issue,” speculation abounds.

On theory holds that “flash trades” like Knights resulted from the well-known “warm-up phase” required by Java Virtual Machine (JVM) technology. In a nutshell, JVM solutions enable Java to run on any platform, powering a host of enterprise applications including trading platforms. The “warm-up phase” is the time it takes a JVM solution, and therefore the applications that depend on it, to reach optimal performance levels. Traditionally, JVM solutions are “warmed-up” using test cases, as is explained in such freely-available resources as Oracle’s JRockit JVM technical literature.

In practice, financial services firms frequently warm-up applications using mock, or simulated, trading data. The risk, of course, is that a chain of events could cause that mock data to be considered live and touch off a cascade of unintended trades. According to Paul Rowady, Senior Analyst for the TABB Group, any known technology issue that creates trading risk, such as the JVM warm-up, is worth investigating. “In my Wall Street & Technology Case Study Feature JVM Warm Up Phase– Pg 2 of 2 experience with high-frequency trading, it makes sense to consider the impact of ‘warming-up’ JVM,” he says. While Rowady stresses he’s unaware of the exact nature of Knight’s issue or the relative magnitude of risk that warming-up JVM poses, he does believe it’s an issue that many highly-automated trading businesses face. “If there’s a viable, cost-effective solution, firms may want to check the issue off the list and move on to something else, given the enormity and complexity of the unknowns,” he says.

Unsurprisingly, at least one technology firm has announced a solution. Sunnyvale, Calif.-based Azul Systems recently launched ReadyNow!, a proprietary warm-up mitigation technology now exclusively available within its own JVM solution, called Zing. Beyond reducing the risk, Azul points out that insufficiently warmed systems may not operate at optimal performance at market open due to the dynamics involved in real-world trading environments. With the introduction of ReadyNow!, Azul says Zing provides “peak application performance” without extended warm-up periods. Whether large Wall Street firms with significant investments in established JVMs, such as Oracle’s JRockit or HotSpot (obtained in the Sun Microsystems acquisition), jump ship to adopt a solution like Zing remains to be seen. However, for those firms considering the adoption of a new trading platform, or even upgrading existing systems, investigating new JVM technologies presents an opportunity. All other things being equal, deploying a platform without the warm-up risk certainly can provide competitive advantages. Anne Rawland Gabriel is a technology writer and marketing communications consultant based in the Minneapolis/St. Paul metro area. Among other projects, she's a regular contributor to UBM Tech's Bank Systems & Technology, Insurance & Technology and Wall Street & Technology ... View Full Bio

Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
Anne R. Gabriel
50%
50%
Anne R. Gabriel,
User Rank: Apprentice
4/24/2014 | 4:00:17 PM
re: Did JVM's "Warm-Up" Cause Knight's Collapse?
Thanks, Ivy! Agreed that stock exchange kill switches are a necessary safeguard. Hopefully there will eventually be a public discussion of the actual source of the Knight issue. Regardless, the JVM warm-up delay is definitely an issue in a real-time HFT world and critical for trading firms to examine and mitigate.
IvySchmerken
50%
50%
IvySchmerken,
User Rank: Author
4/23/2014 | 4:16:44 PM
re: Did JVM's "Warm-Up" Cause Knight's Collapse?
Very interesting piece about the "JVM warm-up risk" that could have triggered incidents like Knight's "cascade of unintended trades."
You would think that nearly two years later, that someone would have publicly discussed this as a case study to explain what happened so that other firms can avoid this risk. Even though the Nasdaq kill switch is not getting at the underlying cause, it seems to be a necessary safeguard.
Anne R. Gabriel
50%
50%
Anne R. Gabriel,
User Rank: Apprentice
4/21/2014 | 10:36:26 PM
re: Did JVM's "Warm-Up" Cause Knight's Collapse?
Thanks, Becca! The delay is in minutes as systems are brought up to speed before market open. During that time, "dummy" trades are used for JVM to "learn" what it needs to do. Obviously, if these dummy trades get into the market (and set off corresponding trades), that's a big problem.

After market open, once systems are "warm" if JVM encounters a different situation than what occurred during warm-up (a different type of trade than was part of the dummy data), then there's a pause of milliseconds while JVM adjusts to running this new type of trade.

Hence, there are two types of opportunities for havoc to occur as a result of this known JVM behavior.
Becca L
50%
50%
Becca L,
User Rank: Author
4/21/2014 | 2:03:27 PM
re: Did JVM's "Warm-Up" Cause Knight's Collapse?
Great piece! I had never heard of this warm-up risk before. Not knowing anything about JVM it does seem possible the delay can pose a real threat to a firm's trades (and thus the broader market). Do you know if this warm-up delay a matter of seconds? Minutes? Milliseconds?

We've been learning time and time again that sometimes the biggest errors don't exclusively come from the elaborate schemes of disgruntled employees and malicious hackers, it can come from the simplest oversights. The Heartbleed hack might also be an example of that, too.
Register for Wall Street & Technology Newsletters
White Papers
Current Issue
Wall Street & Technology - Elite 8, October 2014
The in-depth profiles of this year's Elite 8 honorees focus on leadership, talent recruitment, big data, analytics, mobile, and more.
Video
Exclusive: Inside the GETCO Execution Services Trading Floor
Exclusive: Inside the GETCO Execution Services Trading Floor
Advanced Trading takes you on an exclusive tour of the New York trading floor of GETCO Execution Services, the solutions arm of GETCO.