Author: Michael Walsh, Enterprise Architect
As we advance nearer to January of 2018, the topic of MIFID II compliance has become very prevalent in my meetings with trading firm executives. Specifically, the RTS 25 (time stamping requirements) is a frequent point of discussion. Although it has recently been defined as a technical standard, precise time stamping is not a new requirement.
For context, when I started in this industry many years ago, I was what’s known as a ‘runner’ on the CME trading floor. A runner’s job was to take the paper confirmations of executed trades from the locals or brokers, and literally run the cards to a large, heavy machine that looked like a combination between a safe and a stapler. This machine had rolling digit dials that would leave an ink impression of the current time, when you moved paper over an internal switch. As a runner, I stamped each trade card by at least two of these devices within 1 minute of each other; which was the timestamping requirement of the time.
I chuckle when I take into consideration that my annual salary in those days was roughly equivalent to the cost of a switch capable of the same job today! If the analogy of the runner has helped you conceptualize the requirement, read on and I will further explain the rules and tools that need to be understood to comply with this standard.
RTS 25 Requirements
When boiled down, there are few demands in this standard for electronic market participants:
- Synchronize with Coordinated Universal Time (UTC) from a recognized time source.
- Document the system design with evidence of consistent latencies throughout.
- Compare the local clock to another official time source periodically to validate that the clock is calibrated.
If you receive an acknowledgement from the exchange for a new order in 1 millisecond or less, your timestamps must be in 1 microsecond granularity, with no more than 100 microsecond variance from UTC. Otherwise, all other electronic trades need to be recorded at 1 millisecond accuracy.
Clock Synchronization Solutions
Although any computer it capable of maintaining these levels of precision for a short while, their clocks will gradually drift out of compliance unless you synchronize with an officially recognized time service. In general, there are two options:
National Laboratory Network
The most sophisticated science laboratories in the world are the custodians of extremely accurate atomic clocks. Atomic clocks set their time off of trace amounts of radioactive material. The rate of decay of these unstable atoms is so predictable, that a clock set to this entropy is unsurpassed in accuracy. Fiber optic networks that connect to these laboratories can deliver this time as a service to their customers.
GPS and other national satellite networks are the most widely used time sourcing technology used today. The satellites used for time synchronization via GPS are also atomic clocks. They operate exactly as the clocks in the laboratory example above. The clocks themselves are no less accurate than the national laboratories, and their time signal can be received by any GPS antenna from every point on Earth.
Which is better?
You may be asking, “If the satellites have the same accuracy as the atomic clocks on the ground, why would anyone invest in the much higher cost of connecting directly to a ground-based time service?” Well, since you asked… it’s relativity, of course! The satellites are moving faster than the Earth, which is how they are able to stay in orbit perpetually. Time is relative to motion, so faster moving clocks are counting time slower. Now, these satellites are being updated more frequently than their ability to count time as well, so this variance in time accuracy is mitigated continuously, by the same fundamental method as the switch, server, or stamp clock. The loss of accuracy due to these principles of physics could cause a variance of 10 nanoseconds at any given time.
Now, will having a network connection to an atomic clock provide better accuracy than the GPS system? In theory, yes, but keep in mind that the National Laboratory atomic clocks are contributors to UTC. UTC is the weighted average across many national contributors across the globe. It is also the basis for the satellite Global Positioning System (GPS). Therefore, a single nation’s atomic clock could also be 10 nanoseconds divergent than the accepted UTC time, and would be no better in that circumstance than a GPS timestamp.
The most differentiating factor between direct connection to a UTC contributor lab and satellite is the transmission medium. A physical network will have high reliability, however network devices fail and fiber paths are mistakenly cut more often than they should. Although wireless transmission can be temporarily disrupted by natural phenomenon like solar flares, my experience is that the time to recover from this is less than a break in communication from a physical network, and therefore I believe GPS is the better choice.
Choosing a Grandmaster
With the time source decided, we now need a centralized authority for time in your own network. A Grandmaster clock is the term used for the device that owns this role. There are several companies that make Grandmaster/Time Server appliances, for example FSMLabs, Microsemi, or Spectracom, to name a few. Consider it your own personal atomic clock. In fact, that’s exactly what it is, although it is not quite as capable as all of the world’s greatest atomic clocks combined. For this reason, it will rely on the pulsed updates from the GPS antenna, which is mounted to the datacenter’s roof, and directly connected via copper cable. These devices generally do not have many access ports, so the final consideration is what devices you will use for your time distribution network.
To extend the time service to many endpoints, network devices that support the feature can be configured to be what is known as a boundary clock. Boundary clocks are often dedicated to the delivery of a time service. High throughput volumes from other traffic has the potential to cause unpredictable delays in the time updates. As such, many time networks are completely isolated from the common network and run parallel to the core and core distribution network.
DIY, or MSP?
When deconstructed into its parts, RTS 25 should not seem so intimidating. Even at this small scale, however, there is a lot of new technology for a small-to-medium firm to build and maintain. If that thought was at the forefront of your mind as you read through this article, then consider partnering with a managed service provider like NetXpress.
At NetXpress, we designed our local network to deliver RTS 25 compliant time services, raw market data feeds, and datacenter interconnectivity, all over the same network port, which is uncommon amongst MSPs because of the difficulty in doing so. It’s no challenge for NetXpress, however, due to our unique network design. The LAN distribution network is all Layer-1, making for a highly deterministic latency to each endpoint. Our collapsed core architecture reduces not only jitter, but cost and overhead , and the economics of simplicity are passed through to our clients.
Rest easy, knowing that the time service NetXpress clients receive is dead-on. Why? Because we are regularly comparing our local time sources to the Grandmasters hosted at each of our Co-location facilities, with real-time alerting if they skew.
Lee Segall said “A man with a watch knows what time it is. A man with two watches is never sure”. With NetXpress, you can have two watches, and still be sure.
To learn more about this topic, and the other abilities of our service, contact us.