OSADL QA Farm on Real-time of Mainline Linux
About - Hardware - CPUs - Benchmarks - Graphics - Benchmarks - Kernels - Boards/Distros - Latency monitoring - Latency plots - System data - Profiles - Compare - Awards
Effect of sleep state setting on latency and power consumption
Wakeup latency of all systems - Real-time optimization - Peer-to-peer UDP duplex link - OPC UA PubSub over TSN - Powerlink - Ethercat - Network load - kvm - Sleep states
For testing purposes, particular sleep states were disabled on one of the 12-way systems (6 cores with hyperthreading). Core #2 was not allowed to enter any sleep state, and core #3 was not allowed to enter a sleep state deeper than 1. Core #1 was left unmodified, i.e. any sleep state can be entered according to the system load. In addition, the scaling governor of CPU #2 was set to performance from 5.05 a.m. to 5.05 p.m. and left at ondemand (same as the other cores) in the remaining time period. For details, please refer to the related profile section.
Last update 6 minutes ago
The data clearly demonstrate that the sleep state setting has no effect on the latency and on the power consumption while the 200-µs cyclic activity of the cyclictest program is creating continuous system load. Under idle conditions, however, only the core that is not allowed to enter a sleep state deeper than 1 continues to deliver the same real-time capabilities. The other cores suffer from up to 10 times higher latencies. On the other hand, allowing cores to enter a deep sleep state has a clear beneficial effect on the power consumption. Note that the power consumption reflects the entire system and not only the processor.