streamich / usockets

Miniscule cross-platform eventing, networking & crypto for async applications

µSockets - miniscule networking & eventing

This is the cross-platform async networking and eventing foundation library used by µWebSockets.

It runs on Linux, macOS, FreeBSD and Windows. Most is C11 while some optional parts are C++17.

Key aspects

• Built-in (optionally available) TLS support exposed with identical interface as for TCP.
• Acknowledges and integrates with any event-loop via a layered hierarchical design of plugins.
• Extremely pedantic about user space memory footprint and designed to perform as good as can be.
• Designed from scratch to map well to user space TCP stacks or other experimental platforms.
• Low resolution timer system ideal for performant tracking of networking timeouts.
• Minimal yet truly cross-platform, will not emit a billion different platform specific error codes.
• Fully opaque library, inclusion will not completely pollute your global namespace.
• Continuously fuzzed by Google's OSS-Fuzz with 95% code coverage (asan, ubsan and msan).

Extensible

Designed in layers of abstraction where any one layer depends only on the previous one. Write plugins and swap things out with compiler flags as you see fit.

Compilation

Build example binaries using make examples. The static library itself builds with make. It is also possible to simply include the src folder in your project as it is standard C11. Defining LIBUS_NO_SSL (-DLIBUS_NO_SSL) will disable OpenSSL 1.1+ support/dependency (not needed if building with shipped Makefile). Build with environment variables set as shown below to configure for specific needs.

Available plugins

• Build using WITH_LIBUV=1 make [examples] to use libuv as event-loop.
• Build using WITH_GCD=1 make [examples] to use Grand Central Dispatch/CoreFoundation as event-loop (slower).
• Build using WITH_OPENSSL=1 make [examples] to enable and link OpenSSL 1.1+ support (or BoringSSL).
• Build using WITH_WOLFSSL=1 make [examples] to enable and link WolfSSL 2.4.0+ support for embedded use.

The default event-loop is native epoll on Linux, native kqueue on macOS and finally libuv on Windows.

A word on performance

This library is opaque; there are function calls for everything - even simple things like accessing the "user data" of a socket. In other words, static linking and link-time-optimizations mean everything for performance. I've benchmarked dynamic linking vs. static, link-time optimization and found roughly a 60% performance difference.

The kernel you run makes a huge difference. Linux wins, hands down. Mitigations off, or a modern hardware-mitigated CPU makes huge differences and distros like Clear Linux have shown significant speedups compared to more "vanilla" kernels.