ARCHITECTURAL CONCEPTS

NVMe High-Speed Data Capture and Recorder

The Data Capture and Recording Challenge

There are many streaming data sources customers want to capture or record, but achieving this at high speed (100+ Gb/s) is challenging—that is where we can help!

What sensor inputs are possible?

Any sensor data that can be transported via ethernet frames can be captured and recorded by BittWare’s Data Capture and Recorder architectural concepts.

This can include video, RF or virtually any other sensor or network data.

Do you need analog conversion to or from digital? Our RFX-8440 and RFX-8441 cards might be an excellent fit!

What do we mean by Data Capture?

For Data Capture (concepts #1 and #2 below), sensor data passes through the FPGA and CPU to host DDR4 DRAM memory.

The capture depth is typically limited to seconds for high data rates. DDR4 DRAM is relatively expensive.

What do we mean by Data Recorder?

Processed or raw sensor data can also be recorded on disk. We present several recorder architecture concepts as shown below.

Concept #1: Capture

Overview

The smallest and least complex architectural approach is to capture sensor data straight to host memory. Software on the host schedules the data for processing by multiple CPU cores.

Click here to learn more about Data Capture in our white paper.

Key Points

  • Quickly take advantage of PCIe Gen 4 and 5 as they emerge
  • The smallest and least complex architecture
  • Capture buffer size limited to host memory capacity
  • Some BittWare cards have support for PCIe expansion that can add capture bandwidth

Concept #2: Capture and Record

Overview

The simplest architecture to record at high bandwidth, this approach extends the Data Capture of concept #1 by enabling the data captured in host memory to be streamed out to storage devices for recording

Key Points

  •  Capture & Record data beyond the limits of host DDR memory capacity

  • No limit to the number of NVMe drives – enables easy capacity expansion

  • Maximum recording rate a function of the storage array’s maximum streaming write bandwidth

Concept #3: Direct Record to Disk

Overview

In this architecture, the FPGA PCIe card streams data into the NVMe array on the same PCIe root complex as the CPU. This allows data flow to bypass host memory under software control while minimizing CPU impact.

Key Points

  • Avoids the complexity and expense of FPGA direct-attached NVMe

  • Enable the use of an external NVMe JBOD box connected to the host through a PCIe expansion card

  • Leverages Linux p2pmem (newer kernels)

Concept #4: Record to FPGA Direct-Attached Disk

Overview

In this architecture, NVMe drives/backplanes are directly attached to a BittWare FPGA card via cables. The host can still access the drives, but with the FPGA acting as storage controller.

Key Points

  • NVMe drives can be connected to the FPGA directly on BittWare cards with compatible expansion ports
  • Maximum recording rate is a function of the storage array’s maximum streaming write bandwidth

Concept #5: NVMe Over Fabric Recorder

Overview

In this architecture, the FPGA sends the sensor data directly to a network-attached commercial NAS server via NVMe-oF.

Key Points

  • Allows use of commercial NAS Servers

  • This concept leverages NVMe-oF inside the FPGA

  • Most commercial NAS servers are limited to under 100 Gb/s for sustained line rates

  • This concept takes the most resources inside the FPGA

Want More Details?

Data Capture White Paper and App Note

Learn more about 100G Data Capture with our white paper and download the detailed App Note describing its operation. For BittWare customers with Xilinx-based products, you can also download the Capture project on the Developer website.

Ready to Learn More?

Details on our NVMe High-Speed Data Capture and Recorder are available upon request. Get in touch with us to learn more!