Information Technology Reference
In-Depth Information
Performance Terminology
In Chapter 1 we defined some performance-related terms we will use throughout this
chapter and the rest of the topic; we summarize those terms here.
Task. A user request. A task, sometimes called a job, can be any size, from skat-
ing the mouse across the screen to computing the shape of a newly discovered
protein. When discussing scheduling, we use the term task, rather than thread or
process, because a single thread may be responsible for multiple user requests,
e.g., in a word processor, each character typed is an individual user request to
add that character to the file and display the result on the screen.
Response time (or delay). The user-perceived time to do some task.
Predictability. Low variance in response times for repeated requests.
Throughput. The rate at which tasks are completed.
Scheduling overhead. The time to switch from one task to another.
Fairness. Equality in the number and timeliness of resources given to each task.
Starvation. The lack of progress for one task, due to resources given to a higher
priority task.
In this chapter, we will try to give you the conceptual and analytic tools to
help you answer these questions.
The key topics in this chapter are:
Uniprocessor scheduling. How do uniprocessor scheduling policies af-
fect fairness, response time and throughput?
Multiprocessor scheduling. How do scheduling policies change when
we have multiple processor cores per computer?
Energy-ecient and deadline scheduling. Many new computer sys-
tems allow energy to be saved by running computations more slowly. How
do we make this tradeoff, while still minimizing the impact on user per-
ceived response time? More generally, how do we make sure our jobs finish
in time?
Queueing theory. In a server environment, how are response time and
throughput affected by the rate at which requests arrive for processing
and by the scheduling policy?
Overload control. How do we keep response time reasonable when a
system becomes overloaded?
