Skip to main content
Software Engineering Daily

ByteDance’s Container Networking Stack with Chen Tang

47 min episode · 2 min read
·

Episode

47 min

Read time

2 min

AI-Generated Summary

Key Takeaways

  • eBPF kernel programming: Developers write C programs that compile to bytecode, pass through a safety verifier, then load into the Linux kernel without modules or restarts, enabling dynamic packet filtering and system tracing in production environments.
  • Hardware offloading strategy: ByteDance combines eBPF with smart NIC hardware by using a slow path-fast path separation where eBPF processes initial packets, then an agent translates rules to hardware that caches them for thirty-second intervals, bypassing kernel overhead.
  • Container networking at scale: Traditional Kubernetes service discovery becomes a bottleneck above 100,000 machines because indexing all backend containers creates unacceptable overhead, requiring ByteDance to build custom service discovery frameworks that operate without global state management.
  • RDMA integration technique: eBPF enables RDMA direct memory access for containers by first identifying destination locations through kernel hooks, then passing connectivity information to NICs that can bypass kernel stack entirely for subsequent packets between containerized applications.

What It Covers

ByteDance engineer Chen Tang explains how the company uses eBPF technology to manage container networking across over one million servers, replacing traditional virtual switches with kernel-level packet routing for improved efficiency and scalability.

Key Questions Answered

  • eBPF kernel programming: Developers write C programs that compile to bytecode, pass through a safety verifier, then load into the Linux kernel without modules or restarts, enabling dynamic packet filtering and system tracing in production environments.
  • Hardware offloading strategy: ByteDance combines eBPF with smart NIC hardware by using a slow path-fast path separation where eBPF processes initial packets, then an agent translates rules to hardware that caches them for thirty-second intervals, bypassing kernel overhead.
  • Container networking at scale: Traditional Kubernetes service discovery becomes a bottleneck above 100,000 machines because indexing all backend containers creates unacceptable overhead, requiring ByteDance to build custom service discovery frameworks that operate without global state management.
  • RDMA integration technique: eBPF enables RDMA direct memory access for containers by first identifying destination locations through kernel hooks, then passing connectivity information to NICs that can bypass kernel stack entirely for subsequent packets between containerized applications.

Notable Moment

Chen reveals that ByteDance can inject observability code directly into the kernel of live production containers, collect diagnostic data from specific function calls and contexts, then remove the tracing program without any system restart or service interruption.

Know someone who'd find this useful?

You just read a 3-minute summary of a 44-minute episode.

Get Software Engineering Daily summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

More from Software Engineering Daily

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

This podcast is featured in Best Cybersecurity Podcasts (2026) — ranked and reviewed with AI summaries.

You're clearly into Software Engineering Daily.

Every Monday, we deliver AI summaries of the latest episodes from Software Engineering Daily and 192+ other podcasts. Free for up to 3 shows.

Start My Monday Digest

No credit card · Unsubscribe anytime