RBS: Profile-Guided Scheduling for Time-Triggered Applications

Time-Triggered networks like Time-Triggered Ethernet, TTP/C, Flexray, and IEEE 802.1Qbv are used in critical embedded systems to provide reliable and predictable message delivery. These networks work by controlling the maximum size and exact rate at which applications can transmit messages. While this approach works well for applications that generate fixed-sized messages at constant rates, it leads to inefficient bandwidth utilization when message sizes and timing are highly variable. For example, to accommodate an application that runs up to X times per second, and generates up to Y bytes of data per execution, the network needs to be scheduled for the worst-case — a Y -byte message every 1/X seconds.We introduce Reduced Bandwidth Scheduling (RBS), a novel method for reducing the bandwidth allocated to highly-variable applications while still ensuring they meet reliability requirements. RBS models an application’s communication pattern as a biased Bernoulli distribution, which we prove represents the worst-case message overhead. With this model, RBS can provide an upper bound on the probability that an application overflows its network buffers, and a lower bound on message reliability. Our evaluation shows that, for representative embedded applications, RBS can reduce bandwidth utilization by up to 329 kbps (43% of the original utilization) while requiring only 2000 bytes of extra buffer space. Moreover, in a realistic case study with a representative audio compression application, RBS reduced bandwidth utilization by 146 kbps (34% of the original utilization) while requiring only 41040 bytes of extra buffer space (20× more) and maintaining reliability greater than 0.999 over 15 years.