Skip to content

tests/eas/acceptance: Reduce expected LITTLE busy time for offload test

Darryl Green requested to merge github/fork/bjackman/reduce-offload-expected-busy-time into master

Created by: bjackman

Here's a trace plot that shows an example of what I'm talking about in the commit message (note that the blue tasks are placed on the same CPU when the 'migrator' tasks start running, which reduces the busy time. IIUC although this isn't ideal behaviour it is not unexpected.):

offload_failure

Where the 'migrator' tasks in the OffloadMigrationAndIdlePull are initially placed on the same CPU, the busy time of the LITTLE CPUs that are exercised can be below 99% during the perdiod where big CPUs are occupied. This shouldn't be considered a failure.

In my experiments 97% appeared to be a low enough value to prevent these failures. In fact I observed no failures with the value at 98% either, but it seems wise to allow a little margin for error.

Merge request reports