Hierarchical CBS design

Development @https://github.com/jlelli/linux/issues/4

Comparison w.r.t. RT_GROUP_SCHED

FEATURE	DEADLINE	RT
affinity	any, since RT entities can freely migrate/be pinned to any (v)CPU of the group	any
RT_SHARE	tasks migrate	runtime migrates
hierarchy	only leaf groups can contain tasks; are intermediate nodes RT or DL?	no restriction (might cause starvation!)
scheduling	EDF at root level/FP between tasks	FP at root level/FP between tasks
throttling	only with groups (need to see how to include root)	at all levels (including root)
bandwidth	reserved on each CPU (even if not used, might be reclaimed with GRUB)	used only if tasks are running

Questions

• What to do with RT_RUNTIME_SHARE?
  • make tasks migrate to CPUs (servers) with available bandwidth
  • what if however a task has restricted affinity and can't migrate?
• Hierachy
  • EDF only at root level? Can't be useful to have EDF also between subgroups (e.g., a user that gets a fraction of bandwidth and wants to further divide it)
  • tasks only on leaf nodes?
  • how to enable throttling at root group level as well? maybe create default (e.g. 95%) servers?
• Unused bandwidth
  • use GRUB to let tasks/groups reclaim leftover bandwidth?