Preview Session
summary
Preview session summarizes concrete claims and tools across NSDI: RDMA vs kernel‑bypass shows one‑sided RDMA accesses can skip remote CPU but require NIC co‑processor tradeoffs; smartNICs and FPGA prototypes (TONIC, NetTLP) target line‑rate transport offload; NVMe/NVMM papers (FileMR, i10) build file‑backed RDMA and in‑kernel NVMe stacks and report near‑RDMA throughput; programmable packet scheduling reduces hardware state via limited priority approximation; measurements use atlas/scans and looking‑glasses; security demos include homomorphic/time‑encrypted queries, enclave hardening and PrivateEye VM compromise detection.
AI-written summary · about
chapters (47)
- 0:00Introduction
- 2:39Agenda
- 3:06Networking
- 5:54Network Layers
- 10:38DPDK vs RDMA
- 11:30RDMA
- 11:55Onesided Networking
- 12:27Coprocessor Limitations
- 13:23Polling
- 13:53Building Applications
- 14:25Wrapping Up
- 14:59Optimizations
- 15:24Programmable Packet Scheduling
- 17:07Conclusion
- 17:42Next Presenter
- 18:43Data Center
- 23:35Network Measurements
- 26:43Session Overview
- 27:55Demand for Memory
- 28:13Moores Law
- 28:31CPU Cycles
- 28:52Smartnics
- 29:12ASICbased Smartnic
- 29:31FPGAbased Smartnic
- 29:50SOCbased Smartnic
- 30:21Excel TCP
- 31:28Tonic
- 32:38NVM
- 33:47NVME
- 36:24NetTLP
- 36:38What is PCIe
- 38:06Pando
- 38:28Goals
- 39:33Summary
- 41:37Data Center Network
- 42:01Monitoring Infrastructure
- 44:58Monitoring Reporting
- 47:47Internet Measurement
- 50:54Reproducibility
- 51:16Progress
- 52:22Techniques
- 52:46Control Plane
- 54:01Data Plane
- 54:54Network Telescopes
- 55:45Papers
- 56:06Layer 8 Techniques
- 58:44Wrapup
related talks
02h 53mPreview Sessions
12mPreview: Networking
14mUnderstanding RDMA Microarchitecture Resources for Performance Isolation
17mWhite-Boxing RDMA with Packet-Granular Software Control
18mRDMA is Turing complete, we just did not know it yet!
10mDemo: SmartNIC with OvS-DPDK on RISC-V
15mCharacterizing Off-path SmartNIC for Accelerating Distributed Systems
15mSecuring RDMA for High-Performance Datacenter Storage Systems
