Dive deep into the performance and tuning aspects of service mesh data planes in this 38-minute conference talk from CNCF. Explore the latency challenges associated with sidecar introduction in service mesh adoption and understand the performance breakdown from kernel to user space. Learn about innovative approaches implemented at Tencent that achieved a 50% latency improvement, including the use of eBPF SOCKMAP and Envoy task scheduler extensions. Gain insights into kernel acceleration techniques, Envoy latency analysis, Istio Mixer telemetry optimization, and future developments in service mesh technology. Discover practical solutions to enhance service mesh performance and overcome adoption barriers in cloud-native environments.
Deep Dive into Service Mesh Data Plane Performance and Tuning
CNCF [Cloud Native Computing Foundation] via YouTube
Overview
Syllabus
Intro
What is Service Mesh?
Istio Mesh
Latency concerns in Istio 1.3
Performance Comparison Across Version
Latency Hotspot
Is It All about the Kernel?
What is eBPF?
How does eBPF Help?
Kernel Acceleration with eBPF
What else we can do?
About Envoy
Measure Envoy Latency
Envoy Latency Data Analysis
Let's Dig Deeper
Istio Mixer Telemetry Optimization
Mixer Telemetry Optimization Result
Further thinking
Istiod Evaluation
Wasm-based Telemetry V2
Envoy Wasm Performance Disadvantage (1)
tstats Telemetry Filter
tstats Optimize Metric Calculation
tstats Optimization Results
Future Works
Taught by
CNCF [Cloud Native Computing Foundation]
