Note 1: This repository contains the code for FlueNT10G's hardware design, as well as the PCAP import/export and agent software. The Go software library for programmable control over the network tester is located at https://github.com/aoeldemann/gofluent10g.

Note 2: The measurement applications and evaluation scripts referenced in the Reproducible Research section of our FPL 2018 paper titled FlueNT10G: A Programmable FPGA-based Network Tester for Multi-10-Gigabit Ethernet are available here.

What is FlueNT10G?

FlueNT10G is an open-source, free-to-use FPGA-based network tester. FlueNT10G can precisely replay pre-recorded or synthetically generated network traces, capture incoming network traffic and accurately record network latencies on a per-packet basis.

Where is the documentation located?

The documentation can be found at https://fluent10g.readthedocs.io. Although it is still in its early stages, we are continuously extending it.

Are there usage examples?

Yep! Please have a look in the examples folder of this repository.

What replay/capture throughput does FlueNT10G achieve?

FlueNT10G can concurrently replay and capture network traffic on three 10 Gbps network interfaces for all packet sizes. When operated exclusively in replay or capture mode, four 10 Gbps network interface can be served.

How precise is the replay of network traces?

FlueNT10G replays traffic according to the timestamps included in the trace file. The 10G MAC IP cores are operated with a clock frequency of 156.25 MHz. Thus, the achievable replay precision is 6.4 nanoseconds.

How accurate are traffic capturing and latency measurements?

FlueNT10G records the arrival time (and optionally the latency) of each packet it receives. Since the 10G MAC IP cores are operated with a clock frequency of 156.25 MHz, the timestamping precision of our design is 6.4 nanoseconds. However, we found that the 10G MAC IP core on Xilinx Virtex-7 devices introduces minor delay variations on the receive-side. In our experiments we observed a timestamp precision of 32 nanoseconds. We plan to port our design to an UltraScale FPGA board in the future to achieve even higher accuracy.

Is there a maximum trace file size?

Instead of pre-loading trace data to the memory of the FPGA board before replay can start, FlueNT10G continuously streams data between host computer and the FPGA board during active measurements. Thus, the maximum trace and capture file size is only limited by the memory capacity of the host computer.

What do I need to use FlueNT10G?

FlueNT10G is composed of an FPGA-based hardware component, as well as software running on a standard PC. Currently our implementation targets the NetFPGA-SUME board. However, if you are familiar with FPGA hardware development, the effort to port our design to another board should be modest. To generate the FPGA bitstream, you need a license for the Xilinx Vivado tool chain. Additionally, a license for the Xilinx 10 Gigabit Ethernet MAC IP core is required. To sum up, you will need:

• NetFPGA-SUME or (with some development effort) another PCIe Gen3 FPGA board
• Standard PC with a free PCIe Gen3 slot
• License for Xilinx Vivado Design Suite
• License for Xilinx 10 Gigabit Ethernet MAC IP core

Where can I find more information?

FlueNT10G was presented at the International Conference on Field Programmable Logic & Applications 2018 in Dublin, Ireland. Please use Google Scholar to find a digital copy of our paper. BibTeX:

@inproceedings{fluent10g-fpl2018,
  author={{Oeldemann, Andreas and Wild, Thomas and Herkersdorf, Andreas}},
  title={{FlueNT10G: A Programmable FPGA-based Network Tester for Multi-10-Gigabit Ethernet}},
  booktitle={Field Programmable Logic and Applications (FPL), 28th International Conference on},
  year={2018}
}

Which Xilinx Vivado version should I use to build the FPGA bitstream?

FlueNT10G has been tested to build with the Xilinx Vivado 2018.3 release.

Who is behind FlueNT10G? How can I get in touch with the team?

FlueNT10G is actively being developed by a team of researchers at the Chair of Integrated Systems at the Technical University of Munich. Please contact Andreas Oeldemann in case of questions, feedback, problems and so on.