Glossary
- averaging window
- A sliding window, 100 ms long by default, over which the thread scheduler calculates the CPU percentage usage.
- bankruptcy
- What happens when critical threads exhaust their partition's critical time budget.
- budget
- The CPU time, expressed as a fraction of 100%, that a partition is guaranteed to receive when it demands it.
- CPU share
- Another word for budget.
- critical budget
- A time, in milliseconds, that critical threads are allowed to run even if their partition is out of CPU budget.
- critical thread
- A thread that's allowed to run, even if its partition is out of CPU budget, provided its partition has a nonzero critical budget.
- fair-share schedulers
- A class of thread schedulers that consider dynamic processor loads, rather than only fixed thread priorities, in order to guarantee groups of threads some kind of minimum service.
- free time
- A time period when some partitions aren't demanding their guaranteed CPU percentage.
- inheritance
- What happens when one thread, usually a message receiver, temporarily adopts the properties of another thread, usually the message sender.
- inheritance of partition
- What occurs when a message-receiving thread runs in the partition of its message-sender.
- microbilling
- Calculating the small fraction of a clock tick used by threads that block frequently, and counting this time against the threads' partitions.
- partition
- A division of CPU time, memory, file resources, or kernel resources with some policy of minimum guaranteed usage.
- scheduler partition
- A named group of threads with a minimum guaranteed CPU budget.
- thread scheduler
- Lets you guarantee minimum percentages of the CPU's throughput (using budgets) to groups of threads, processes, or applications.
- throttling
- Not running threads in one partition, in favor of running threads in another partition, in order to guarantee each their minimum CPU budgets.
- underload
- The situation when the CPU time that the partitions demand is less than their CPU budgets.