We know that reducing the sections with preemption and IRQ disabled reduces the latency, also that IRQs influences on it, but some cases are hard to catch. For example, in the old jump label update, there was a burst of IPIs causing latency spikes. Such non-periodic behavior is hard to mathematize. As a side effect, this adds pessimism to "possible formulas" that tries to define the worst-case latency, mainly regarding IRQs. Daniel would like to discuss his idea about the possible approaches to this problem, without adding unpractical pessimism.
|I agree to abide by the anti-harassment policy||Yes|