May 2026 Rustacean meetup

Runtime: 90 minute Technical Talk with QnA

Abstract

For a while now, tokio has been the “default” choice for Rust’s async ecosystem, popularized by its robust work-stealing scheduler. However, as high-performance networking and hardware locality become central to the conversation, the thread-per-core (shared-nothing) architecture is gaining significant traction.

This talk explores the fundamental design trade-offs between these two philosophies. We will try to understand:

• Scheduler Mechanics: How work-stealing manages load balancing versus how thread-per-core minimizes context switching and cache misses.
• The API Tax: Why thread-per-core runtimes often require !Send futures and how that reshapes application architecture.
• Operational Reality: Choosing a runtime based on your workload. Is your workload CPU or I/O bound?

Key Takeaways

1. Overview of the internal machinery of the tokio scheduler vs. io_uring optimized runtimes.
2. Identifying how runtime choice dictates the use of synchronization primitives (e.g., Arc<Mutex<T>> vs. local state).
3. A framework to help you understand when to stick with the “standard” and when to pivot to a specialized architecture.