SYSC 5701 Operating System Methods for RealTime Applications

1
SYSC 5701Operating System Methods for Real-Time
Applications

• Scheduling Aperiodic and Sporadic Jobs in
  Priority Driven Systems
• Winter 2009

2
Liu Ch. 7

• Assumptions
• now allow aperiodic or sporadic tasks!!
• tasks are independent (still)
• will relax assumption later
• interdependency ? Liu Ch. 8
• uniprocessor

3
Aperiodic Events

• aperiod?ic a. having no natural frequency
• stochastic inter-arrival times
• some may be critical sporadic period
• sporad?ic a. intermittent scattered single
• will consider techniques
• background
• polling
• sporadic server

4
Aperiodic Service in Background

• service aperiodic tasks as background process
• lowest priority always preempt for periodic
  tasks
• OK for non-critical, less important activities
• may cause aperiodic deadlines to be missed
• could starve aperiodic work

5
Background (cont)

• may cause poor response time for aperiodic work
• response time
• time of completed service time of request
• best case no pending periodic work
• worst case all periodic work pending
• background ? OK for some cases, but

6
Integrate with Periodic?

• how to ensure aperiodic work not starved with RM
  approach?
• must cast aperiodic tasks into periodic framework
• must know task inter-arrival characteristics
• allocate a periodic task that does nothing but
  service aperiodic work!

7
Priority Queues in Kernel
add periodic server jobs to periodic queue
periodic server
periodic jobs
processor
aperiodic jobs
periodic server jobs process aperiodic jobs
8
Aperiodic Service Using Polling

• goal improve response time to aperiodic
  requests
• schedule periodic aperiodic service as a
  regular server task period ps, budget es
• when aperiodic job released
• ? put it in aperiodic job queue
• when aperiodic server executes
• polls aperiodic job queue and execute requests
• wont exceed allocated budget

9
Polling (cont)

• if all pending aperiodic jobs completed before
  consuming budget
• server exhausts remaining budget set 0
• suspends itself
• budget replenished at beginning of each period
• What if job arrives just after server suspends
  self, even though budget existed for job?
• What to do if aperiodic job misses deadline?
• How to schedule aperiodic jobs in aperiodic queue?

10
Bandwidth-Preserving Server

• deferrable server
• server does not exhaust budget before suspending
  self
• budget replenished to es at start of each period
• allows late arrivals to execute without waiting
  until next period
• improves response time of aperiodic jobs

11
Deferred Server

• different replenishment algorithm!
• has sliding window replenishment period
• replenishment period starts when server begins
  execution with a full budget

not in replenishment period
full budget
release
replenishment period
12
Performance ?

• Sprunt, Lehoczky, Sha
• deferred server response time improvement
• 6 times better than polling
• 10 times better than background
• o/s implementation issues
• more complex
• hybrid static dynamic scheduling
• managing server execution capacity
• ? starts to look like time-slicing !?

13
Theory?

• for deferred server delaying lower priority jobs
• worst case critical instant
• server is highest priority ready job
• has almost full budget (es )
• es remaining in period
• ? will be replenished after es

14
Deferred Server Critical Instant
worst case possibility!
t0
t0 t

• over time t from critical instant, server can
  delay lower priority tasks by
• t es

t
t es
es
es
budget from previous period
end of previous period
? ?
? ?
t es
es
es
es
1
ps
ps
15
Equation 7.1 in Liu Text
delay due to higher priority periodic jobs
max. delay due to server
?
i-1
? ?
t es
ek
?i(t) ei es
es
ps
k 0
16
Schedulable Utilization

• Theorem 7.2 in text

for n periodic tasks
?? ? ?
1 (n1)
us 2 us 1
URM/DS(n) ( n 1)
1
es ps
where us
17
Server Priority

• if server priority lt priority of task i
• ? server does not influence response time of task
  i
• if server priority gt priority of task i
• ? use Equation 7.1

18
Multiple Servers?

• might prefer to have different priorities
  associated with different aperiodic tasks
• use a server for each priority level
• if m servers with priority gt priority of task i

?
i-1
( )
? ?
t es,k
ek
?i(t) ei 1
es,k
ps,k
k 0
19
Sporadic Server

• worst case ? replenish server at the end of every
  sporadic server period
• essentially the same as deferred server
• acceptance test?
• only accept if enough slack budget to complete
  job before deadline

20
Fixed Priority Acceptance Test

• assume sporadic jobs placed in sporadic job queue
  in EDF order
• slack ? of a job server budget left over after
  executing the job
• influenced by other jobs accepted by server!

21
Slack of a Job

• for job Si( t , ds,i , es,i )

execution of previously accepted jobs with
earlier deadlines
?
?s,i(t) ds,i t es es,i
(es,k ?s,k )
? ?
ps
ds,k lt ds,i
number of replenishments before deadline ?
conservative !
completed portion of job
server period
server budget
22
Acceptance

• must have enough slack
• accepting wont make previously accepted jobs
  late (wont influence jobs with earlier
  deadlines!)
• ? each job with a later deadline must have at
  least es,i slack at t

23
Priority Queues in Kernel
add periodic server jobs to periodic queue
periodic server
periodic jobs
processor
2
aperiodic jobs
1
sporadic jobs
test
periodic server jobs process aperiodic and
sporadic jobs
reject