YoVDO

FlexTOE - Flexible TCP Offload with Fine-Grained Parallelism

Offered By: USENIX via YouTube

Tags

USENIX Symposium on Networked Systems Design and Implementation (NSDI) Courses Parallelization Courses SmartNICs Courses

Course Description

Overview

Explore a conference talk on FlexTOE, a flexible TCP offload engine for SmartNICs that eliminates host data-path TCP processing while remaining fully customizable. Learn about its focus on data-path offload of established connections, fine-grained parallelization techniques, and modular design for high performance on wimpy SmartNIC architectures. Discover how FlexTOE compares to host TCP stacks and other TOEs, offering improved scalability for Memcached, reduced host CPU usage, and lower latency for RPCs. Understand its support for C and XDP programs written in eBPF, enabling implementation of popular data center transport features. Gain insights into the challenges of TCP offload to SmartNICs, the FlexTOE offload architecture, and the benefits of parallel TCP processing across different hardware platforms.

Syllabus

Intro
High CPU Overhead of TCP
Need for Flexible TCP Offload • FlexibilityDatacenter networks evolve rapidly
TCP Offload to SmartNICs - Challenges
FlexTOE Offload Architecture
Parallelizing the TCP Data-path for Offload
Parallel TCP Processing Example: Transmit (TX)
Evaluation Setup
Benefits of Offload: Throughput Scalability
Is Fine-grained Parallelism Necessary?
Data-path Parallelism: Does it Generalize across Platforms?


Taught by

USENIX

Related Courses

Scaling Memcache at Facebook
USENIX via YouTube
Multi-Person Localization via RF Body Reflections
USENIX via YouTube
Opaque - An Oblivious and Encrypted Distributed Analytics Platform
USENIX via YouTube
Live Video Analytics at Scale with Approximation and Delay-Tolerance
USENIX via YouTube
Clipper - A Low-Latency Online Prediction Serving System
USENIX via YouTube