low latency

You are currently browsing articles tagged low latency.

Market Structure Documentary Comes to the Silver Screen

October 16, 2012 in Current Events, Fi-Tech Culture, Regulations | Permalink

Tired of explaining to friends and family that  many of Wall Street’s latest fiascos aren’t due to conspiracies but to technology?

Wait a few more months and take them to see the market structure documentary that Arbitrage Pictures plans to release next Spring.

The movie, Ghost Exchange, examines what happens what happens when trading speeds and complexity outpace regulatory oversight and understanding.

The filmmakers interview a number of well-known industry members and experts to find whether the widespread adoption of dark pools, high-frequency trading and trading algorithms has placed Wall Street, and the rest of the global markets, at the whims of “the ghost in the machine” that now appears to run the markets.

Check out the online trailer here.

 

 

 

 

 

Tags: , , ,

Proprietary Market Data Feed Design Costs NYSE Euronext $5 Million

September 16, 2012 in Architecture, Regulations | Permalink

Electronic traders know that milliseconds of latency can cost them a lot. Now, NYSE Euronext know that too.

Without admitting or denying allegations by the US Securities and Exchange Commission (SEC) Exchange, NYSE Euronext officials announced on Sept. 14th that the exchange agreed to pay a $5 million penalty to settle latency discrepancies between two of its proprietary market data feeds and the data it delivers to the Consolidated Quote System (CQS).

According to SEC officials, NYSE Euronext violated Rule 603(a) of Regulation NMS, which prohibits exchanges to deliver quote and trade data to its proprietary feed clients before it sends the same data to the consolidated quote feed, as well as the record-retention section of 17(a)(1) of the Securities Exchange Act.

The regulator inspected several high-volume trading days from early 2010 and found various 15-second intervals where NYSE Euronext’s market data distribution system (MDDS) experienced average delays exceeding 25, 50 or 100 milliseconds, often near the market close, say officials. A number of these periods also had more than 10 percent of their quotes delayed more than a second, including one day in February 2010 where a “substantial” percentage of quotes were delayed by more than five seconds during the last 30 seconds of trading, the officials add.

This would not have been an issue if NYSE Euronext’s Open Book Ultra and PDP Quote, also known as NYSE BBO, feeds also reflected the same delays. However, NYSE Euronext staff designed the exchange’s proprietary feeds to run in parallel with its MDDS. So when the MDDS experienced its delays, the proprietary feeds did not.

NYSE Euronext officials attribute the timing differences to technology issues that they have resolved.

“The violations at the NYSE may have been technological, but they are not technical,” said Daniel Hawke, chief of the SEC’s Division of Enforcement in a prepared statement. “Robust technology governance is just as important to preventing investor harm as any other compliance or supervisory function.”

As part of the settlement, NYSE Euronext will hire an outside consultant, at its own expense, to compile a report on the exchange operator’s current Rule 603(a) compliance of its NYSE, NYSE Arca and NYSE MKT MDDS platforms as well as the related internal policies and procedures. The consultant will then have a chance to recommend changes to improve policies, procedures and platforms, which NYSE Euronext will need to implement.

This action by the SEC is a clear signal to the other regulated exchange operators to have their own proprietary market data feeds up to snuff in regards to Rule 603(a). Although $5 million might be a rounding error to many investment banks, it is not for exchanges. What’s worse is the potential remediation costs they face if they are in breach of the rule.

The SEC probably would have taken an exchange out to the woodshed earlier on this matter if it was not for all of the time and it has spent in rolling out Dodd-Frank regulations with the US Commodity Futures Trading Commission (CFTC) over the past couple of years. But now it sounds like mommy is definitely home.

 

Tags: , , ,

You May Be a High Frequency Trader If…

June 22, 2012 in Regulations, Trading | No comments

Earlier this week, one of the Technology Advisory Committee (TAC) working committees presented the US Commodities Futures Trading Commission (CFTC) a proposed four-point definition of high-frequency trading.

It’s important to remember that this definition is nothing more than a recommendation to the CFTC. As CFTC Commissioner Scott O’Malia stated publicly in the past, the working group’s definition will be the straw-man definition that will spur industry discussion and lead to an eventual regulatory definition.

According to the working group members, their goal was to keep the definition broad enough to capture any future practices within the existing definition and avoid narrow language that might lead to regulatory arbitrage.

Given the rise of high frequency trading has forced more firms to focus on market data and trade messaging latency as well as algorithmic execution like never before, how do you know whether your strategy would be considered high frequency trading under the proposed rule?

Here is definition as the working group presented it to the CFTC:

 

High frequency trading is a form of automated trading that employs:    

(a) algorithms for decision making, order initiation, generation, routing, or execution, for each individual transaction without human direction;

(b) low-latency technology that is designed to minimize response time, including proximity and co-location services;

(c) high-speed connections to markets for order entry; and

(d) high message rates (orders, quotes or cancellations).

 

Most people in the industry would agree that all four points accurately describe high frequency trading, but it still leaves the regulators and the industry the challenge of turning a subjective concept into an objective definition. What qualifies as low latency technology, high-speed connections and high message rates?

Any fixed figure would be outdated almost immediately, which leaves regulators using fixed ratios to define high frequency trading. Yet the management, documentation and reporting of staying between those regulatory lines would be prohibitive to all but the largest firms with the deepest IT pockets.

The TAC has managed to grab the low-hanging fruit with its recommendation, now the hard work really begins.

Tags: , , , ,

Zeptonics Splits Switch in Half to Lower Latency

June 20, 2012 in Architecture, Trading | No comments

Although most of my discussions during this week’s SIFMA Tech Leadership Forum and Expo center around middle- and back-office operations, I had a quick meeting with Australian network equipment provider Zeptonics, which announced the second its 50-port low-latency Layer-1 ZeptoLink networking device, which can be paired with its 24-port 10 Gbps Ethernet layer-2 ZeptoMux multiplexing switch that it announced a few weeks ago

According to David Snowdon, a hardware principal at Zeptonics, high-performance layer-2 switches offer similar performance whether the traffic is “fanning up” from multiple links into a single exchange link or “fanning down” from an exchange link to multiple consumers.

To address fanning-up performance, Zeptonics built the ZeptoMux using FPGAs that provides a reported fibre-to-fibre latency of about 130 nanoseconds.

To improve the fanning-down performance, the company decided to use ASICs rather than FPGAs in the ZepoLink and bit streaming rather than packet streaming, which delivers a latency of approximately 5 nanosecond.  The ZepoLink also acts as an intelligent patch panel allowing multiple 1-to-1 and 1-to-n connection within the same device.

Zeptonics should have 50 ZeptoMux units to beta clients in the coming weeks and plans to deliver ZeptoLink units to beta users sometime in the third quarter.

 

Tags: , , , ,

Chalk Up Another Win for Low Latency Appliances

June 18, 2012 in Architecture, Disruptive Technology, Trading | No comments

I first met NovaSparks CEO Yves Charles a couple of years ago sitting around a tiny conference room in my old office.

I  remember the meeting clearly because we has about six or seven people sitting around a small table designed for about three people. The other thing I remember from the meeting was the introduction to NovaSparks’ low-latency market data appliance.

Using FPGAs for low-latency messaging is not a new concept and several vendors and investment banks offer FPGA-based products that you could slip into your co-located servers. However, the NovaSparks engineers decided to create a standalone FPGA-based appliance that runs market data feed handlers for the leading global cash equities markets directly on the FPGA cards without accessing an external CPU.

During the discussion, we spoke about a number of functions that could be added to the platform eventually.

Last week one of those ideas came to fruition as NovaSparks official announced that the platform could build an entire order book on the appliance and send the normalized to the consuming downstream applications either via multiple 10 Gbps Ethernet or PCI-e connections, depending on the appliance’s configuration.

I haven’t seen the appliance used in anger yet, but according to Yves, it can build the order book in less than 1.5 microseconds 99.99 percent of the time and to just under a microsecond for 99 percent of the time. Or as he likes to say, “Most of the time, it will average around 880 nanoseconds.”

There must be something to these performance numbers since low-latency execution broker Lime Brokerage officials announced today that they plan to deliver ultra-low latency market data feeds to their clients via the NovaSparks platform.

 

Tags: , , , , ,

« Older entries