birq-1.6.0 and birq-1.5.1

Added by Serj Kalichev over 1 year ago

The birq-1.5.1 is bugfix release. The command line options "--threshold" and "--load-limit" were fixed (same options in config file were ok). The 1.5 branch has a mess with command line options, config file options and config re-read by SIGHUP. So it's recommended to use new birq-1.6.0.

The birq-1.6.0 has new config file option "use-cpus". It's a affinity mask. It can coexist with older "exclude-cpus" option. The real affinity will be 'affinity = <use-cpus> & ~<exclude_cpus>'. The command line options for IRQ processing (like strategy, threshold, intervals etc.) will be removed. That options can be defined by the config file only now.


Added by Serj Kalichev over 4 years ago

The config file is implemented. The config file is preferred method to configure birq daemon. The command line options to set parameters is a legacy method. Use SIGHUP to re-read config file.

The 'exclude-cpus=<cpumap>' option is implemented. It allows to exclude some CPUs from the list of CPUs that process IRQs.


Added by Serj Kalichev over 6 years ago

Conservative birq behaviour. The birq on start tries to link IRQs to CPU using current affinity. It doesn't change affinity if IRQ doesn't have affinity to multiply CPUs. It doesn't change IRQ affinity if number of interrupts is 0.

In some cases the OS can't move IRQ to specified CPU. To keep the state consistent the birq analyzes real IRQ affinity on each iteration.

Add new option --load-limit. If the loading of CPU is greater than this limit then no IRQs can't be moved to this CPU. It stabilizes the behaviour on a high load.

Version 1.1.1

Added by Serj Kalichev almost 8 years ago

Sometimes while a really high load the previous version of birq can mistakenly blacklist IRQs on start. It was fixed in birq-1.1.1. Also some warnings were fixed.

Version 1.1.0

Added by Serj Kalichev almost 8 years ago

The proximity configuration file is implemented. It's useful for platforms with broken NUMA node proximity. Using this configuration file, the user can set PCI device proximity to a specific NUMA node or to a subset of CPUs (CPU mask).


Also available in: Atom