Linux Now Lets You Skip Syscalls by Turning Processes Into Libraries — And It’s Blazingly Fast
From 19,000 CPU cycles down to 1,700. Is that even legal?
Engineers at ByteDance have introduced a groundbreaking method to accelerate interprocess communication (IPC) in Linux, dubbed RPAL — Run Process As Library. The idea is bold yet elegant: instead of treating processes as separate entities that talk via syscalls, RPAL lets one process call another like a regular in-memory function, bypassing the kernel entirely. The result? Dramatically lower latency and reduced CPU overhead.
Traditional IPC mechanisms in Linux — like sockets, pipes, or other syscall-based interfaces — often become performance bottlenecks in high-load systems. RPAL takes a different approach: minimize memory copying, avoid kernel context switches, and do it all without major changes to application code. Essentially, RPAL turns a standalone process into a dynamically linkable library.
Early benchmarks from ByteDance show impressive results. In one test, a client process sent 1 million 32-byte messages. With standard IPC, each message consumed about 19,616 CPU cycles. With RPAL, that dropped to 1,703 cycles — a 91% reduction.
Faster, Leaner, Smarter IPC
In real-world datacenter scenarios, RPAL reportedly reduced CPU usage by up to 15.5%, thanks to fewer syscalls and better memory sharing. A key optimization was shared address space between processes, which cut down on data duplication and memory overhead.
But there's a catch: RPAL relies on Memory Protection Keys (MPK) — a hardware feature available on modern Intel and AMD Zen 4+ CPUs. ByteDance notes that MPK-free support is on the roadmap, but not available yet.
The company has already published RFCs and initial kernel patches, making RPAL open for community review and experimentation. If adopted widely, it could mark a major shift in Linux IPC models, especially for high-performance microservices and real-time systems.
Would you like a technical breakdown of how RPAL works under the hood?
