During this season of giving, you can show your support for the NTP Project by making a donation to Network Time Foundation.

High-Throughput Low-Jitter Stratum 1 NTP Server (GSoC 2009)


Highly accurate synchronization of PC and server clocks is crucial for many applications. Microprocessor based NTP servers suffer from a large amount of timestamp jitter (e.g., ~50000 ns), for a variety of reasons such as OS scheduling and unpredictable bus timing. The goal of this project is to implement a stratum-1 NTP server in reconfigurable logic, capable of gigabit Ethernet linerates with less than 8 ns of jitter, and a flexible stratum-0 interface.


Linerate throughput and low jitter will be achieved by implementing a custom datapath for NTP requests and timestamping logic in reconfigurable logic. A custom NTP datapath interfacing with the physical layer Ethernet transceiver means the receive and transmit timestamps can be accurate to within the clock cycle (8 ns with gigabit Ethernet). Furthermore, as NTP and ARP requests and replies are symmetrical in size, a datapath responding only to these packets can never overflow transmit FIFO's in extreme situations, and the appliance itself is immune to DDoS attacks. As no software is involved, the system can also easily be verified as being crack proof.

To accommodate auxiliary functions such as management and logging, the system will also feature an embedded microprocessor and a secondary Ethernet interface.

The project has three components: the hardware appliance, firmware (custom logic), and software.


Stratum-1 NTP servers implemented on COTS (commodity off the self) microprocessor systems suffer from a large amount of timestamping jitter due to the non-deterministic latency during processing and as data is moved through the system.

A stratum-1 NTP server implemented with reconfigurable logic has the following advantages over a microprocessor based implementation:
  • Less physical space required;
  • Lower power consumption;
  • Immune to DDoS attacks (although the network may still be susceptible);
  • Crack proof;
  • Line rate throughput;
  • Timestamp jitter < clock cycle period (8 ns for GMII).


The project started as a summer project at the WAND Network Research group in December 2007.


The hardware schematics will be captured with gEDA gschem, and layed out with gEDA's PCB. Pre-layout simulations will be performed with Mentor Hyperlynx.

Rather than use a "soft" microprocessor such as the microblaze and a device such as the Spartan-3, we are now tending towards the more expensive Virtex-4 with its "hard" PowerPC 405 micro and dual tri-mode MACs. A tentative list of features for the board is:

The board is expected to be 6 - 10 layers.

Appliance Hardware Estimates has more information.


The logic will be described with VHDL. Verification will be performed with Mentor Modelsim.

Firmware for a PRS10 rubidium clock and either GPS or ACTS disciplining will be implemented during the summer of code.


Date Task Description % Done
Accepted student proposals announced on the Google Summer of Code 2009 site.
Community Bonding
Students get to know mentors, read documentation, get up to speed to begin working on their projects.
Coding begins
Students begin coding for their GSoC projects
Start mid-term evaluations
Mentors and students can begin submitting mid-term evaluations.
Mid-term evaluations
Mid-term evaluations deadline
Suggested 'pencils down' date. Take a week to scrub code, write tests, improve documentation, etc.
Firm 'pencils down' date.
Mentors, students and organization administrators can begin submitting final evaluations to Google.
Final evaluation
Final evaluation deadline
Code samples
Students can begin submitting required code samples to Google


Hardware Appliance

  • Schematics;
  • Layout design;
  • Gerber plots;
  • Fab drawings.


  • Source code (VHDL);
  • Synthesized bitstream.


  • Source code (mostly C);
  • Binary image.


  • Architectural specification;
  • Component documentation.

NICT (Japan) have reported a hardware stratum-1 NTP server implemented on a PCI card with a reconfigurable logic device: High Performance NTP Server using FPGA. In constrast to the NICT implementation, this project will not require a host PC and will be open source.

Future Work

Future work includes support for IEEE1588 and IPv6. Both of these will be possible on the hardware platform with the large reconfigurable logic device chosen. IPv6 may be implemented during the summer of code if time allows.

Discussion and Comments

Topic revision: r17 - 05 Oct 2022, DruLavigne
Copyright © by the contributing authors.Use of this website indicates your agreement with, and acceptance of, the PrivacyPolicy, the WikiDisclaimer, and the PrivateWebPolicy.