Multiprocessor

Real-Time Scheduling
Embedded System Software Design
Use under the permission of Prof. Ya-
Shu Chen (NTUST)
 Outline
• Multiprocessor Real-Time Scheduling
• Global Scheduling
• Partitioned Scheduling
• Semi-partitioned Scheduling
 Multiprocessor Models
• Identical (Homogeneous): All the processors have the same
characteristics, i.e., the execution time of a job is
independent on the processor it is executed.
• Uniform: Each processor has its own speed, i.e., the
execution time of a job on a processor is proportional to the
speed of the processor.
– A faster processor always executes a job faster than slow
processors do.
– For example, multiprocessors with the same instruction set but
with different supply voltages/frequencies.
• Unrelated (Heterogeneous): Each job has its own execution
time on a specified processor
– A job might be executed faster on a processor, but other jobs
might be slower on that processor.
– For example, multiprocessors with different instruction sets.
 Scheduling Models
• Global Scheduling:
– A job may execute on any processor.
– The system maintains a global ready queue.
– Execute the M highest-priority jobs in the ready queue, where M
is the number of processors.
– It requires high on-line overhead.
• Partitioned Scheduling:
– Each task is assigned on a dedicated processor.
– Schedulability is done individually on each processor.
– It requires no additional on-line overhead.
• Semi-partitioned Scheduling:
– Adopt task partitioning first and reserve time slots (bandwidths)
for tasks that allow migration.
– It requires some on-line overhead.
 Scheduling Models
Semi-partitioned
Global Scheduling Partitioned Scheduling Scheduling

7
9 8 7 3 7 7
Waiting Queue
4 6 5 4 6 5
6 4 5
1 2 3 1 2 3

2 3

CPU 1 CPU 2 CPU 3 CPU 1 CPU 2 CPU 3 CPU 1 CPU 2 CPU 3

 Global Scheduling
• All ready tasks are kept in a global queue
• A job can be migrated to any processor.
• Priority-based global scheduling:
– Among the jobs in the global queue, the M highest priority
jobs are chosen to be executed on M processors.
– Task migration here is assumed with no overhead.
• Global-EDF: When a job finishes or arrives to the global
queue, the M jobs in the queue with the shortest
absolute deadlines are chosen to be executed on M
processors.
• Global-RM: When a job finishes or arrives to the global
queue, the M jobs in the queue with the highest
priorities are chosen to be executed on M processors.
 Global Scheduling
• Advantages:
– Effective utilization of processing resources (if it works)
– Unused processor time can easily be reclaimed at run-
time (mixture of hard and soft RT tasks to optimize
resource utilization)
• Disadvantages:
– Adding processors and reducing computation times and
other parameters can actually decrease optimal
performance in some scenarios!
– Poor resource utilization for hard timing constraints
– Few results from single-processor scheduling can be
used
 Schedule Anomaly
• Anomaly 1
A decrease in processor demand from higher-
priority tasks can increase the interference on a
lower-priority task because of the change in the time
when the tasks execute
• Anomaly 2
A decrease in processor demand of a task negatively
affects the task itself because the change in the task
arrival times make it suffer more interference
Anomaly 1
Anomaly 2
 Dhall effect
• Dhall effect : For Global-EDF or Global-RM, the least
upper bound for schedulability analysis is at most 1.
• On 2 processors:
Task T D C U

T1 10 10 5 0.5

T2 10 10 5 0.5

T3 12 12 8 0.67

• T3 is not schedulable
 Schedulability Test
• A set of periodic tasks t1, t2, . . . , tN with
implicit deadlines is schedulable on M
processors by using preemptive Global EDF
scheduling if

where tk is the task with the largest utilization

Ck/Tk
 Weakness of Global Scheduling
• Migration overhead
• Schedule Anomaly
 Partitioned Scheduling
• Two steps:
– Determine a mapping of tasks to processors
– Perform run-time single-processor scheduling
•Partitioned with EDF
– Assign tasks to the processors such that no
processor’s capacity is exceeded (utilization
bounded by 1.0)
– Schedule each processor using EDF
 Bin-packing Problem

The problem is NP-complete !!

 Bin-packing to
Multiprocessor Scheduling
• The problem concerns packing objects of
varying sizes in boxes (”bins”) with the
objective of minimizing number of used boxes.
– Solutions (Heuristics): First Fit
• Application to multiprocessor systems:
– Bins are represented by processors and objects by
tasks.
– The decision whether a processor is ”full” or not is
derived from a utilization-based schedulability test.
 Partitioned Scheduling
• Advantages:
– Most techniques for single-processor scheduling
are also applicable here
• Partitioning of tasks can be automated
– Solving a bin-packing algorithm
• Disadvantages:
– Cannot exploit/share all unused processor time
– May have very low utilization, bounded by 50%
 Partitioned Scheduling Problem
Given a set of tasks with arbitrary deadlines, the
objective is to decide a feasible task assignment
onto M processors such that all the tasks meet
their timing constraints, where Ci is the
execution time of task ti on any processor m.
 Partitioned Algorithm
• First-Fit: choose the one with the smallest
index
• Best-Fit: choose the one with the maximal
utilization
• Worst-Fit: choose the one with the minimal
utilization
 Partitioned Example
• 0.2 -> 0.6 -> 0.4 -> 0.7 -> 0.1 -> 0.3

0.1 0.1 0.3

0.3
0.3 0.7
0.6 0.6 0.1
0.7 0.7 0.6
0.4 0.4 0.4
0.2 0.2 0.2

First Fit Best Fit Worst Fit

EDF with First Fit
 Schedulability Test
Lopez [3] proves that the worst-case achievable utilization for
EDF scheduling and FF allocation (EDF-FF) takes the value

If all the tasks have an utilization factor C/T under a value α, where
m is the number of processors

 m +1
EDF − FF
U wc (m,  ) =
 + 1 where  = 1/  
 Demand Bound Function
• Define demand bound function as

• We need approximation to enforce polynomial-time

schedulability test
Deadline Monotonic Partition
Schedulabiliy Test
 Weakness of Partitioned Scheduling
• Restricting a task on a processor reduces the
schedulability
• Restricting a task on a processor makes the
problem NP-hard
• Example: Suppose that there are M processors
and M + 1 tasks with the same period T and
the (worst-case) execution times of all these
M + 1 tasks are T/2 + e with e > 0
– With partitioned scheduling, it is not schedulable
 Semi-partitioned Scheduling
• Tasks are first partitioned into processor.
• To reduce the utilization, we again pick the
processor with the minimum task utilization
• If a task cannot fit into the picked processor, we
will have to split it into multiple (two or even
more) parts.
• If ti is split and assigned to a processor m and the
utilization on processor m after assigning ti is at
most U(scheduler,N), then ti is so far schedulable.
 Semi-partitioned EDF
• Tmin is the minimum period among all the tasks.
• By a user-designed parameter k, we divide time into slots
with length S = Tmin/k .
• We can use the first-fit approach by splitting a task into 2
subtasks, in which one is executed on processor m and the
other is executed on processor m + 1.
• Execution of a split task is only possible in the reserved
time window in the time slot.
• Applying first-fit algorithm, by taking SEP as the upper
bound of utilization on a processor.
• If a task does not fit, split this task into two subtasks and
allocate a new processor, one is assigned on the processor
under consideration, and the other is assigned on the
newly allocated processor.
 Semi-partitioned EDF
• For each time slot, we will reserve two parts.

If a task ti is split, the task can A processor can host two split
be served only within these two tasks, ti and tj. ti is served at
pre-defined time slots with length the beginning of the time slot,
xi and yi . and tj is served at the end.

The schedule is EDF, but if a split task instance is in the ready queue, it is
executed in the reserved time region.
 Semi-partitioned EDF
• We can assign all the tasks ti with Ui > SEP on a dedicated
processor. So, we only consider tasks with Ui no larger SEP.

When executing, the reservation to serve ti is to set

xi to S X (f + lo_split(ti )) and yi to S X (f + high_split(ti )).
SEP is set as a constant.
 Two Split Tasks on a Processor
• For split tasks to be schedulable, the following sufficient
conditions have to be satisfied
– lo_split(ti ) + f + high_split(ti ) + f <= 1 for any split task ti .
– lo_split(tj ) + f + high_split(ti ) + f <= 1 when ti and tj are assigned on
the same processor.
• Therefore, the “magic value” SEP

• However, we still have to guarantee the schedulability of the

non-split tasks. It can be shown that the sufficient condition is
Schedulability Test
Magic Values: f
Magic Values: SEP
 Reference
• Multiprocessor Real-Time Scheduling
– Dr. Jian-Jia Chen: Multiprocessor Scheduling. Karlsruhe Institute
of Technology (KIT): 2011-2012
• Global Scheduling
– Sanjoy K. Baruah: Techniques for Multiprocessor Global
Schedulability Analysis. RTSS 2007: 119-128
• Partitioned Scheduling
– Sanjoy K. Baruah, Nathan Fisher: The Partitioned Multiprocessor
Scheduling of Sporadic Task Systems. RTSS 2005: 321-329
• Semi-partitioned Scheduling
– Björn Andersson, Konstantinos Bletsas: Sporadic Multiprocessor
Scheduling with Few Preemptions. ECRTS 2008: 243-252
See You Next Week
 Critical Instants?
• The analysis for uniprocessor scheduling is
based on the gold critical instant theorem.
• Synchronous release of events does not lead
to the critical instant for global multiprocessor
scheduling