Open Source Summit Japan & Automotive Linux Summit 2021

Linux system calls have generally been implemented as a synchronous mechanism, where a special processor instruction is used to yield userspace execution to the kernel. On system intensive workloads, synchronous system calls negatively affect performance in a significant way, primarily because of pipeline flushing and pollution of key processor structures (e.g., TLB, data and instruction caches). In this talk, we propose the adaptive improvements over typical I/O system calls with fewer mode switches and increased locality of execution in both user and kernel space, that is beneficial to HTTP daemons and database servers. With small modifications against Linux kernel, the new syscall for event-driven scenario and accompanying user-mode runtime translate the synchronous system calls into schedulable ones, and for the application side, it is still more asynchronous than calling I/O calls. It also makes multi-core execution for event-driven programs easier, given that a single user-mode execution context can generate enough requests to keep multiple processors/cores busy with kernel execution. We will show the increase the throughput of Nginx by up to 12%, lighttpd by up to 35% and Redis by up to 3% while only modify two lines of application code.