Data Streaming from SDR to Servers for Cognitive Radar and EW
summary
Shows how to stream multi‑GHz SDR data into GPU servers with concrete numbers and hardware choices: 2 GHz bandwidth at 2.5 GS/s 16‑bit ADC yields ~80 Gbps per channel (160 Gbps for two), kernel networking causes a 10x memory‑bandwidth multiplier and interrupt storms, and kernel bypass with DPDK + SPDK and UHD bypass reduces copies. Recommends direct‑connect quad‑port NICs, balanced PCIe GPU servers (example Supermicro 4124GS for ~16–32 GB/s, 8049U for extreme scale), and contrasts AMD (more PCIe Gen4 lanes) vs Intel for sockets.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:52Need for Cognition in Radar and EW systems
- 2:40Challenges with Cognitive Research Applications
- 4:20Need for High-Channel Count, Heterogenous Compute System
- 6:12Switch and Server
- 11:26Direct Connect
- 12:33DPDK Core Affinity
- 14:17Memory Bandwidth
- 19:12Dual Socket Server
- 25:41Intel Xeon 2nd Generation
- 26:27Dual Socket Epye Server
- 27:39Quad Socket Xeon Server
related talks
22mData transport in a radio telescope: Remote Direct Memory Access over Ethernet from FPGA to GPU
31mHow to capture 100G Ethernet traffic at wire speed to local disk
31mNI: New Capabilities for Comms, Radar, & EW with an SDR-based Research Platform
34mCan the Linux networking stack be used with very high speed applications?
19mProgramming the Electromagnetic Spectrum
15mLive Q&A with Speakers – Joseph Provine
25mKeynote
17mWhite-Boxing RDMA with Packet-Granular Software Control