China Mobile's Panji Platform - Observability Practices and Implementations for LLM Applications

Explore how China Mobile's Panji platform addresses observability challenges in large language model (LLM) applications through this 33-minute conference talk from CNCF. Learn about the complex architectural challenges that LLM deployments present for business observability, including how traditional APM probes consume system resources and impact performance through instrumentation requirements. Discover China Mobile's innovative collaboration with DeepFlow to implement zero-intrusion and full-stack observability using eBPF and WebAssembly (Wasm) technologies. Understand how eBPF enables real-time data collection at the kernel level while Wasm plugins parse streaming requests without code modifications. Examine the platform's integration capabilities that provide service universal mapping, distributed tracing, and multi-dimensional metric analysis to ensure stability and performance optimization of LLM applications. Gain insights into solving coordination challenges when multiple teams manage different LLM instances and learn practical approaches to unified observability construction in complex cloud-native environments.