Task scheduling What are the goals of a

Task scheduling What are the goals of a modern operating system scheduler, and how does Linux achieve them?

Types of scheduling • Stallings identifies three distinct kinds of process-scheduling decisions: – Long-term: which tasks will the system admit? When? And in what order? – Medium-term: which tasks will be temporarily swapped out to disk? And when will they be swapped back in to main memory? – Short-term: which of the ready-to-run tasks will next gain control of the processor?

When are decisions made? • Linux makes its short-term scheduling decisions: – – When a timer interrupt occurs When an I/O request completes When a system-call is invoked When a signal is sent • Linux makes its longer-term scheduling decisions: – When a task exits – When CPU’s idle-time exceeds a given threshold

Goals • Specific scheduling policies are chosen to support desired system behaviors • There are multiple goals -- and sometimes they may even appear to be contradictory • So scheduling policies are a “compromise”

User-oriented goals • • Rapid response-time Short turnaround-times Assured deadlines Predictable performance

System-oriented goals • • • Optimum throughput Maximum CPU utilization Balanced resource allocation Enforce priorities Assure Fairness

Fairness algorithms • One fairness principle is known as FCFS (First-Come, First-Served), though it may not provide optimal throughput • Another way to implement “fairness” is “round-robin” scheduling (“timeslicing”) in which every task gets allocated an equalsize “slice” of the CPUs available time, and all tasks take their turn at executing

Task types • Some tasks are “CPU bound” – They regularly consume the entire amount of processor time that they are allotted • Some tasks are “I/O bound” – They seldom use up their entire timeslice, but instead sleep while awaiting an I/O request

How does kernel distinguish? • If the scheduler repeatedly gets invoked because a task has used up its timeslice, the kernel treats that task as “CPU bound” • If the scheduler repeatedly gets invoked because a task is going to sleep (i. e. , it’s awaiting completion of an I/O request), then that task is treated as “I/O bound”

Responsiveness • To achieve improved responsiveness in interactive applications, an OS kernel can assign a higher priority to I/O bound tasks • Tasks that have higher priority will always get scheduled before any tasks that have lower priority get scheduled • However this could result in “starvation”

Dynamic priorities • Linux combines priority-based scheduling with round-robin scheduling • Linux allows priorities to be dynamically recomputed • Separate queues are used for tasks with differing priorities, while tasks that have equal priority are scheduled round-robin • Tasks can migrate between priorities

In-class exercises • Exercise #1: write an application program which exhibits “CPU bound” behavior • Exercise #2: write an application program which exhibits “I/O bound” behavior • Exercise #3: write a program which would exhibit alternating behavior (becoming an I/O bound task for awhile, then becoming a CPU bound task for awhile)