Seconds matter. And so milliseconds.
Measuring precise time is easier for a brokerage with IT that runs like an Omega Speedmaster chronograph but is a challenge for smaller brokerages with systems that run more like a knockoff Rolex bought outside a bodega. So when the Financial Industry Regulatory Authority proposed a 95 percent improvement broker server-clock synchronization with the atomic clocks of the National Institute of Standards and Technology (NIST), the quartz-sporting brokers stood up and noticed.
Currently, FINRA Rule 7430 allows a one-second drift window between NIST time, the official U.S. time standard, and the clocks in the servers that brokers use to submit mandated reports, such as Order Audit Trail System (OATS) reports. That may sound good but the regulator wants to shrink that window to 50 milliseconds (ms), or five-hundredth of a seconds. Why the time squeeze? To improve market-surveillance quality as well as prepare for the technological foundation of the Consolidated Audit Trail (CAT).
Automated systems have evolved to the point where order placement and trading decisions are made in milliseconds, or fine, basis, FINRA officials wrote in the November 2014 request-for-comment proposal. In such an environment the one-second tolerance is insufficient for audit trails and surveillance periods, particularly since firms will begin trade reporting milliseconds in the near future.
So long, one second tolerance. We don;t have time for you. So far, there are no plans to tighten the existing one-second drift window for mechanical time-stamping devices, the officials added.
Milliseconds are not as short as they used to be, agrees Victor Yodaiken, CEO of server-clock synchronization platform provider FSMLabs. Computers make things faster and I dont think people have come to grips on how necessary it is to have precise timekeeping to keep everything straight in a high-speed trading environment.
Even if FINRA tightened its synchronization window to 100 ms (0.1 seconds) or 200 ms (0.2 seconds), a small scale brokerage like Crew & Associates of Little Rock likely would not meet the proposed performance requirements although it meets the current one-second requirement.
It is extremely unlikely any system change, no matter how expensive, would allow us to remain synchronized to NIST time with less than 200 ms of drift, said Gary Chambers, senior vice president compliance at Crews& Associates.
That Damn Internet
Those wanting to synchronize their server clocks to NIST time can rely on NISTs radio service that broadcasts the time from its facilities in Fort Collins, CO and outside Kekaha, HI or its Internet Time Service (ITS), which it launched in 1993. Both services are free.
Its hard to determine how many people use the radio broadcast since there is no way of knowing who is listening, said Judah Levine, a fellow in NISTs Time and Frequency Division. Our rough estimate is that there are several million users.
Yet, ITS regularly handles 10 billion requests daily, or 10,000 requests per second, over the public Internet, he added.
Crew & Associates Chamber sees the packet-switched nature of the Internet — where network traffic seldom takes the same route twice — introduces too much latency that is out of the participants control to meet FINRAs proposed drift windows without physically re-locating the firms servers.
The network delay results are determined from how far our Internet connection is from the NIST core Internet backbone, which cannot be changed without physically re-locating our firms center of operations, explained Chambers.
These network delays between industry participants and NIST servers can be 200 ms or more, which means Chamber brokerage would only be able to meet the 50 ms-drift window threshold about 90 percent of the time, he adds.
However, NISTs Levine takes exception to this characterization of ITS performance and stated that achieving a 10 ms accuracy is easy.
The accuracy of the request packet when it leaves NISTs servers starts out at the microsecond-level, according to Levine. The millisecond accuracy comes from uncertainty in the path delay and the number of routes and gateways the packet needs to transverse.
To reduce network-based latency, NIST load balances ITS requests among 27 geographically diverse servers across the United States, with the closet server to Wall Street being in CenturyLink Technology Solutions NJ2 Weehawken, NJ datacenter.
Since ITS servers only see a requests network address and not where it originated physically, however, not all requests may reach the nearest ITS server, according to Levine.
NIST attempts load balance ITS workload geographically by posting ITS server IP addresses and locations on the organizations website so that users may select the server physically closet to them.
We push people in that direction, but not everyone does it, admitted Levine.
Routing timing data across wide-area networks (WAN) does not mean maintaining a 50 ms-drift window is impossible, according FSMLabss Yodaiken.
We have clients using our TimeKeeper platform to move time between Sydney and Seoul and keeping it within a 100 ms over not particularly great networks, he claimed. Therealso are firms that have dark-fiber links between Chicago and New York and are keeping their clocks synchronized in the nanosecond range.
Even without additional software or hardware, firms can get within the millisecond accuracy that FINRA proposes by just pointing a computer to a couple of ITS servers, he added. Many of our customers, however, want more than that.
NIST is willing to work with other organizations to improve the delivery of ITS data, including deploying ITS servers in new computing facilities if requested as long as the public could access the servers.
If it was to be accessed only by a specific set of users then we likely would have to charge them for it, said Levine.
Although no one has approached NIST yet for leased-line access to ITS servers, it is in the midst of developing a one that would provide a direct connection to ITS servers without relying on the Internet.
NIST already has performed a few preliminary tests between its Boulder, CO headquarters and a network-switching facility in Florida.
The test was a bit artificial since it was between two network switching facilities and not a customers premises, said Levine, adding that it was too early to discuss possible pricing for the service.
But What If…
As with any regulatory request for comment, FINRAs proposal raises more questions than it originally intended.
NISTs Levine finds the proposal extremely vague from a technical perspective and leaves him wondering if the 50 ms-drift window is meant for average or worst-case performance.
My reading of the document, it has me believe that they meant worst case, but that is not what they say, he said. They do not say anything.
Levine, who was unaware of the proposed rule before being interviewed for this story, suggests that FINRA and Wall Street speak with NIST to help the industry meet any new time-synchronization requirements.
The FINRA proposal is weak, agrees FSMLabs Yodaiken. But, it is a good step forward.
For Crews & Associates Chambers, who finds the proposal overly broad for the industry, he suggest that any changes to drift window tolerances should be limited to only a subset of brokerages.
A separate rule for high-frequency trading would be advisable, Chambers told Traders. We do not engage in such activities and can easily accommodate the current one-second rule.
Chambers also suggests if that the regulator moves forward with an HFT-only requirement, that is updates its auditing rules so that cleared trade timestamps and order timestamps use the one-second threshold for firms that do not engage in HFT.
Given the order are initiated at our firm, requiring the clearing firm to achieve less than a 50 ms-drift for our trades but our inability to achieve the same may result in auditor confusion, Chambers explained.
FINRA could eliminate much of the network-latency issues if it permitted firms to synchronize their server clocks to other atomic clocks operated and maintained by the U.S. government.
Checking In With GPS
The most popular alternative already in use is the Global Position System, which uses its own atomic clocks since its deployment in 1980.
The GPS atomic clocks theoretically are accurate up to 14 nanoseconds (0.000000014 seconds or 14 billionths of a second) to NIST time.
It might be considered heresy if you were a time physicist, said FSMLabs Yodaiken.But for financial services, it is pretty much the same. Its a very efficient time source, but that is not what FiNRA references in its proposal.
A second alternative would have brokerages synchronize their clocks to the US Naval Observatorys master clock, powered by its on ensemble of atomic clocks.
Basically, we all have the same stuff, said NISTs Levine. For all intents and purposes, the times are equivalent.
