Real Time System –

Lecture 14
BITS Pilani Rekha. A
Pilani Campus
 Real time task scheduling
on multiprocessor and
distributed system
BITS Pilani
Pilani Campus
Task Assignment Algorithm

 Static:
• Utilization Balancing Algorithm
• Next – Fit Algorithm
• Bin packing algorithm

 Dynamic:
• Focused addressing and bidding algorithm
• Buddy algorithm

BITS Pilani, Pilani Campus

Utilization Balancing
Algorithm

• Maintain tasks in a queue

• In increasing order of their utilization

• Remove tasks one by one from queue

• Allocates to the lowest utilized process

BITS Pilani, Pilani Campus

Next – Fit algorithm

• Tasks with utilization are divided into classes with

uniform/similar utilization range.
• According to number of classes the number of processors
are decided.
• Each class is assigned to specific processors.

For n Processor:
n classes of tasks is constructed.
A task belongs to class j, iff
(21/(j+1) -1) < ei/pi <= (21/j -1)

BITS Pilani, Pilani Campus

Next Fit Algorithm

Consider 4 processors
Class 4 (0.41,1)
Class 3(0.26,0.41)
Class 2(0.19, 0.26)
Class 1 (0, 0.19)

BITS Pilani, Pilani Campus

Next Fit algorithm

Class4 (0.41,1)
Class 3(0.26,0.41)
Class 2(0.19, 0.26)
Class 1(0, 0.19)

BITS Pilani, Pilani Campus

Bin packing algorithm

Allocate tasks to processors:

Individual processors are scheduled using EDF

Bin packing algorithm

• First fit algorithm
• Best fit algoritm
• Worst fit algorithm
• First fit decreasing algorithms

BITS Pilani, Pilani Campus

First Fit algorithm

• Tasks with utilization are selected and packed in a bin.

• If bin exceeds 100% on further addition of tasks bin2 is
started to be packed.
• Task Utilizations are always whether they can be packed
fully on previous processors.

BITS Pilani, Pilani Campus

First Fit algorithm

1. After utilization exceeds 1 it changes to processor 2.

2. After utilization exceeds 2 it changes to processor 3.
3. So number of processors required is 3 to accommodate the given task
set

BITS Pilani, Pilani Campus

Bin packing algorithm

Best Fit Algorithm:

• It takes the processors and finds out that processor
among the set of all processors which has the least
capacity and can hold the next task.
Example:
P1 utilization: 0.5
P2 utilization: 0.6
P3 utilization: 0.4
P4 utilization: 0.8
Let a task with e/p = 0.24 arrives . To which processor the
task will be assigned?

BITS Pilani, Pilani Campus

Bin packing algorithm

Worst Fit Algorithm:

• Places the next task onto the processor which has the
highest remaining capacity
Example:
P1 utilization: 0.5
P2 utilization: 0.6
P3 utilization: 0.4
P4 utilization: 0.8
Let a task with e/p = 0.32 arrives . To which processor the
task will be assigned?

BITS Pilani, Pilani Campus

First fit decreasing algorithm

Tasks are sorted in the decreasing order of their utilization.

Tasks are selected one by one:

• Assigned to the bin( Processor) to which it can fit in
( utilization does not exceed 1)
• Assigned to the processor to which it first fits.

BITS Pilani, Pilani Campus

First fit decreasing algorithm

Let us assume that we have a set of these 9 tasks with

their utilization given below . Consider 6 processors.

BITS Pilani, Pilani Campus

First fit decreasing algorithm

1 0.83 0.66 0.5 0.5 0.5 0.33 0.33 0.1666

BITS Pilani, Pilani Campus
Dynamic allocation of Tasks

Required when tasks arrive asynchronously at different

nodes:
• Tasks are assigned to the processors as and when they
arise.

Incur high run time overhead:

• The allocator at every node needs to keep track of the
load position at every other node.

BITS Pilani, Pilani Campus

Focussed Addressing and
Bidding
Every processor maintains two tables:

• Status table

• System load table

BITS Pilani, Pilani Campus

Focussed Addressing and
Bidding
Every processor maintains two tables:

• Status table: Contains information about the tasks which

it has committed to run, including information about the
execution time and periods of the tasks.

• System load table: Latest load information of all other

processors of the system.

BITS Pilani, Pilani Campus

Focussed Addressing and
Bidding
The time Axis is divided into windows:
• An window is an interval of fixed duration.

At the end of each window:

• Examine their status table
• Each processor broadcast the fraction of its computing
power in the next window.

Every Processor on receiving a broadcast:

• Updates the system load table

BITS Pilani, Pilani Campus

Focused addressing and
Bidding
When a task arise at a processor, it checks whether the
task can be processed locally itself.

If it can be processed, it updates its status table

If not, then:
• Examines system load table to look for a processor to
offload the task.

BITS Pilani, Pilani Campus

Request for Bid (RFB)

An overloaded node sends out RFBs:

• To the identified lightly loaded processors(The focussed

processors)

By the time the processor gets RFB:

• The focussed processor might have become overloaded.

BITS Pilani, Pilani Campus

Focused addressing and
Bidding
Specifically two time tbid and toffload are calculated. If
tbid<=toffload, then RFB is sent out.

tbid = (estimated time taken by RFB to reach its destination)

+ (the estimated time taken by the destination to respond
with a bid) + (the estimated time taken to transmit the bid
to the processor);

toffload = (task deadline) ‐ [(current time) + (time to move the

task) + (task execution time)];

○ If tbid ≤ toffload; then RFB is sent out

BITS Pilani, Pilani Campus
Focused addressing and
Bidding

The processor that had sent RFB, waits for bids to arrive or
until a specified time has expired since receiving the
task.

Among the bids received, the processor selects one

processor and transfers the task.

BITS Pilani, Pilani Campus

Focused addressing and
Bidding
 The algorithm incurs high communication overhead:
• Periodic transmission of status messages
• Focussed addressing and Bidding

 Communication overhead largely depends on the

window size selected.

 If the window size is large, communication overhead is

reduced.

 However the information at various processors would be

obsolete.
BITS Pilani, Pilani Campus
Buddy Set Algorithm

• Tries to overcome the high communication overhead of

focused addressing and bidding algorithm.

• Very similar to focussed addressing and bidding

algorithm

• Differs in the manner in which the target processor are

found.

BITS Pilani, Pilani Campus

Buddy Set Algorithm

A processor can be in any of the following two states:

• Under loaded
• Overloaded

No periodic broadcast required:

• Broadcast made only when the status of a processor
changes
• Either from overloaded to under loaded or vice versa.

BITS Pilani, Pilani Campus

Buddy Set Algorithm

• Whenever the status of a processor change, it does not

broadcast this information to all processor

• Limits broadcasts only to a subset of processors called

its Buddy set.

BITS Pilani, Pilani Campus

Design of Buddy Set

Design of the buddy set is an important parameter:

• It should not be too large not too small

In multi-hop networks:
• The buddy set of a processor is typically the processors
that are its immediate neighbours.

BITS Pilani, Pilani Campus

Buddy Set Algorithm

Node state: Threshold- based:

• Underloaded
• Overloaded

Selection policy:
• The tasks that fail admission test at the local node are
considered for transfer.

BITS Pilani, Pilani Campus

Buddy Set algorithm

Information policy:
• Based on buddy set
• When a node makes transition into or out of a state, it
informs its buddies

Transfer policy:
• One of the buddies is chosen as the receiver, based on
the load information available.

BITS Pilani, Pilani Campus

Buddy Set Algorithm

6 7

4
5

2
3

0
1

BITS Pilani, Pilani Campus

Buddy Set Algorithm

6 7

4
5

2
3

0
1

BITS Pilani, Pilani Campus

 Buddy Set Algorithm P 1 2 3 4 5 6 7
0 1 2 4
1 3 5 0
2 6 0 3
3 2 1 7
6 7 4 0 6 5
5 7 4 1
6 4 2 5
4 7 5 3 6
5

2
3

0
1

BITS Pilani, Pilani Campus

 Buddy Set Algorithm P 1 2 3 4 5 6 7
0 1 2 4 3 5 6
1 3 5 0 2 7 4
2 6 0 3 4 7 1
3 2 1 7 6 0 5
6 7 4 0 6 5 1 2 7
5 7 4 1 3 6 0
6 4 2 5
7 0 5 3
4 7 5 3 6 4 1 2
5

2
3

0
1

BITS Pilani, Pilani Campus

 Buddy Set Algorithm P 1 2 3 4 5 6 7
0 1 2 4 3 5 6 7
1 3 5 0 2 7
4 4
7 6
2 6 0 3 4
1 7 1
4 5
3 2 1 7 6
0 0
6 5 4
6 7 4 0 6 5 1
7 2
1 7
2 3
5 7 4 1 3
6 6
0 0
3 2
6 4 2 7
5 0
5 5
3 3
0 1
4 7 5 3 6 4 1
2 2
1 0
5

2
3

0
1

BITS Pilani, Pilani Campus

Buddy Set Algorithm - Issues

 Choosing buddy set

• Too large set: High Communication over head
• Too small set: May not be able to find a suitable receiver
with in a buddy set.
• Topology needs to be taken into account while choosing
the buddy nodes.

 Choice of thresholds:
• Large upper threshold: Lower the rate at which tasks will
be migrated.
• The choice of thresholds depend on size of buddy set,
topology.
BITS Pilani, Pilani Campus
Buddy Set Algorithm - Issues

 Thrashing
• A node X, could be a buddy for several nodes. When
many of these several nodes become overloaded, they
migrate their tasks to the node X and making it
overloaded. This results in further migration of task.

• Therefore, the buddy set should be carefully constructed.

BITS Pilani, Pilani Campus