4.3 KiB
| title | aliases | tags | ||
|---|---|---|---|---|
| 19-cpu-scheduling |
|
why shedule?
- keep cpu busy (cpu utilisation)
- so processed dont hae to spend time waithing for the CPU
- even if cpu is always bust executing processses in different orders can change the average time spent in queue
- cpu burst
- (diagram is old the timing will have changed)
- between burst - waiting for io
- so that interactive processes respond quickly
- time spent waiting in total
- average time between visits to CPU
criteria
- CPU utilisation - % time CPU is busy
- throughput - number of processes completed per time unit
- turnaround time - (for a single process) length of time from submitted to completed
- waiting time - (for a single process) time spent waiting for the CPU
- does include time wating for IO
- response time - (for a signle process) average time from the submission of a request to a process until the first response is produced
Preemption
- 4 situations for scheduling
- a process switches from running to waiting
- a process switches from running to ready (due to interrupt)
- a process switches from wauting to ready (due to comletion of IO)
- a process terminates
- non preemptive - scheduling take splace only under 1 nd 4, which has no choice
- premptive - scheduling can take place under 2 and 3 as well. that means a running process has to give way to the processes with higher priority
if you allow interruption it is preemptive
Scheduling algorithms
Gantt charts
The operation of scheduling algortihm is commonly represented in a Gantt chart
First come first served (FCFS)
simplest method- executes processes in the ready queue on a first come first served basis
- when a proces become ready it is put at the tail of the queue
- when the currently executig process terminates, or waits for IO, the process at the front of the queue is selected next
- processes can come at any time - what if two at the same time? choose process with smallest ID to serve first
non-preemptive
pro
- easy to understand and implement con
- waiting time bad
- convoy effect: lots of small process can get stuck behind a big on
- bad response time
fair but not efficient
Shortest job first (SJF)
choose the job in the queue with the shortest burst time.
- equal burst time are executed FCFS (by process id)
pro
- provably optimal average waiting time
con
- never know in advance what the length of next cpu burst will be
- possibility of long processes never being executed
- starvation of process
- not "fair"
predicting next burst time
machine learning sorta - burst length predicted from previous CPU bursts using exponential average equation
preemption example
- p1 is executing
- p2 arrived with burst time shorter that remainging bust time of P1
- non preemptive - P1 finishes
- preemptive - P2 takes CPU
Priority scheduling
- each process allocated a priotit when it arrives
- cpu is allocated to the process with the highest priority
low number is a higher priority
- non preemptive - run processes until finished
- preemptive - higher priority process should take over CPU when they arrive
SJF is a type of priority scheduling where the priority is based on the estimated next cpu burst
Starvation and aging
starvation occurs when a proces waits indefinitely to be allocated the CPU
aging - as a process waits its priority increases
Round Robin
commonly used today
designed for time sharing systems
- small unit of time (time quantum) if defined
- ready queue is treated as a circular list
- cpu schedular goes round the ready queue, allocating the CPU to each process for a time interval up to 1 time quantum
choosing time quantum
- if infinitely large - we have FCFS
- if very small - we have processor sharing - lots of overhead for switching
