Real-time Linux offers advantages in embedded processing but faces challenges with interrupt latency compared to RTOS. Managing real-time workloads on heterogeneous architectures requires tuning to balance shared resources like CPU, memory, and I/O. This presentation explores how Texas Instruments optimizes their SoCs for Real-Time Linux, addressing these challenges and ensuring performance.
Real-time Linux has several advantages in embedded processing but is traditionally limited in achievable worst-case interrupt latency to a real-time operating system (RTOS). Linux's advantages are mostly around the plethora of available software for everything from security to machine learning inference.
The challenge for real-time workloads in heterogeneous architectures, with the limited I/O, external memory bandwidth and latency, and CPU bandwidth shared between TrustZone applications, GPU stacks, and all manner of ‘firmware’ running on the system at the same time with any one of them potentially degrading performance to unacceptable levels. Tuning a system where Linux must share resources with everything else on the SoC involves generic principles that apply for all embedded Linux processors, and some features leveraging capabilities that are device specific.
In this presentation, we discuss some of the ways Texas Instruments debugs and tunes their heterogeneous SoCs for Real-Time Linux workloads and the impact of background load with quality-of-service capabilities of the system-on-chip.
Facebook
Google
GitHub
Linkedin
