Next-generation network time-servers are FPGA-based -

Next-generation network time-servers are FPGA-based


I don’t know about you, but until now I haven’t really spent a lot of time thinking about, well, time , at least insofar as it applies to things like the Internet.

I'm interested in time in general (see my book review, In Search of Time, by Dan Falk), and I'm certainly interested in time as it pertains to my Cunning Chronograph (see Controlling IoT Devices with Mobile Platforms), it's just that I haven't devoted a lot of time to thinking about time in the context of the Internet.

Of course, I'm jolly glad that my computer knows what time it is and that it displays the current time in the status bar at the bottom of my screen, but I really don’t care if we're off by a second or two, and I vaguely assumed that the rest of the world felt much the same way.

How wrong I was. The reason for my change of heart is that I was just chatting to the folks at Microsemi. Of course, I know that they make and supply a lot of components — my particular interest is in their RTAX Rad-Tolerant FPGAs and their SmartFusion2 SoC FPGAs — but I had no idea that they also design, manufacture, and purvey entire systems.

In particular, I was unaware that Microsemi has sold tens of thousands of Network Time Protocol (NTP) servers. In fact, Microsemi is a leading supplier of timing infrastructure, with ~93% of UTC (Coordinated Universal Time — don’t ask why the abbreviation isn’t CUT) world time contributions.

It turns out that accurate knowledge of time is mandatory for the successful functioning of all modern electronic infrastructures, including Data Centers, Wireline Communications, Cellular Communications, Enterprise Communications, Secure Communications, Power Grid Communications, Financial Exchanges, Seismic Exploration, and… the list goes on.

Take the financial market, for example. In the case of US stock trading requirements, SEC Rule 613 Article VI, Sec. 6.8 states that, effective February 27, 2015, transaction logs must have at least 1 millisecond granularity, while accuracy must not exceed 50 milliseconds. You think that's tight? Well, on September 28, 2015, the European Securities and Markets Authority (ESMA) released new timing requirements for stock trading specifying 1 microsecond log granularity for High Frequency Trading (HFT) transactions with accuracy not to exceed 100 microseconds, and with an implementation deadline of January, 2017.

When they first saw these new regulations, I bet a lot of IT folks were pulling their hair out and thinking “How the xxxx are we going to manage this?” I bet those same IT folks are going to be dancing in the streets when they hear about Microsemi's new NTP servers — the SyncServer S600 & S650.

Front (top) and back (bottom) of the SyncServer S600 (Source: Microsemi)

This new generation of ultra-high-capacity NTP servers boast four GigE ports for networking flexibility; they provide NTP Stratum 1-level performance via GNSS Satellites; they offer holdover accuracy protection using one of Microsemi’s miniaturized atomic clocks (approximately the same size as a classic Zippo lighter); and — since security is a number-one concern when deploying NTP servers today — they boast an extreme security suite that includes Microsemi's Security-Hardened NTP Reflector technology.

“What is an NTP Reflector?” I hear you cry. Well, I'm glad you asked, because this is where Microsemi's SmartFusion2 SoC FPGAs come into play.

Microsemi's FPGA-based NTP Reflector technology (Source: Microsemi)

The FPGAs accept all inbound NTP packets, analyze and authenticate them, and dispatch time-stamped NTP packets, all at full line speed. Any non-NTP traffic and/or malformed NTP packets are automatically dropped. The only things that get through to the CPU are any network port configuration messages.

But wait, there's more, because some deployments mandate the use of dedicated physical timing connections in addition to secure NTP server interfaces.

S650 option slots and optional timing I/O module (Source: Microsemi)

This sort of physical layer timing interface is required for all sorts of things, like processing sensor data from a missile firing to capturing sensor data for underwater oil and gas exploration.

Once again, these optional timing modules are based on SmartFusion2 SoC FPGAs, which allows them to support Microsemi's FlexPort technology. Until now, a variety of physical layer interface modules have been used for different applications; these include Time Code hardware, Sine Wave hardware, and Rate Gen. hardware.

Optional timing I/O modules feature FlexPort Technology (Source: Microsemi)

Now, all of these legacy sub-systems can be replaced by Microsemi's optional timing I/O modules, which provide software-selectable signal input and output and an intuitive web-based user interface control, all of which eliminates the need for costly external distribution units, increases speed of deployment, and facilitates ease of operation, thereby saving both time and money.

I tell you, I didn’t realize that Internet-centric time was so interesting. One thing I do know is that I'll be paying a lot more attention to this sort of thing in the future. Also, I'm wondering if my Cunning Chronograph would benefit from one of Microsemi’s miniaturized atomic clocks (that would certainly give me some bragging rights with the other competitors in our Cunning Chronograph Competition).

Good grief; is that the time? I really must dash. I'll see you later.

9 thoughts on “Next-generation network time-servers are FPGA-based

  1. “Maybe also interesting would be the WhiteRabbit project, though a bit of overkill for the Chronograph. It uses FPGA together with a bit of hardware to give sub nanosecond accuracy over 1Gb fiber. KM3NeT, a neutrino detector I've written the embedded soft

    Log in to Reply
  2. “It is called UTC to avoid upsetting France.nIn French, Coordinated Universal Time (CUT) is Temps Universel Coordonnu00e9 (TUC).nRearranging the acronym to UTC makes it politically correct.nUTC is the same as GMT (Greenwich Mean Time) but GMT is a time

    Log in to Reply
  3. “It makes sense to use NTP to characterise the performance of internet links.nnHowever is you just want really good time (way better than NTP) as a client, then it makes a whole lot more sense to use GPS.nnSub-$50 GPS timing modules can give you 15ns a

    Log in to Reply
  4. “Did I hear White Rabbit? Max, you presented a White Rabbit demo at ESC2014… do you remember The Killer Rabbit of Caerbannog?? ;-)nnI'm afraid the White Rabbit HW is a little bit massive for the Cunnig Chronograph, but I'm pretty sure Max could accommo

    Log in to Reply
  5. “Congrats to KM3NeT! The White Rabbit project leader at CERN sent some time ago that you had accomplished the weirdest White Rabbit deployment ever: 100 km into the sea off the coast of Italy at 3500 meter depth! ;-)nnI plan to attend to this year White

    Log in to Reply

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.