Upload
pragya-arya
View
143
Download
0
Embed Size (px)
DESCRIPTION
Energy power efficiency of embedded real time systems through Dynamic Voltage scheduling.
Citation preview
Presented By: Pragya Arya
Btech(cs) 3rd year
Banasthali University
INTRODUCTION
Energy power efficient real time systems
are systems which are power efficient i.e.
uses less energy for computation and
processing in the specified time which
basically lead to light weighted devices
having longer battery life.
REAL TIME SYSTEMS
A real-time system is any information processing
system which has to respond to externally
generated input stimuli within a finite and
specified period.
Mainly Real time systems are divided in two
categories
Hard real time Systems
Soft real time systems
HARD REAL TIME SYSTEMS
An overrun in response time leads to potential
loss of life and/or big financial damage
Many of these systems are considered to be
safety critical.
Examples: Pacemakers , Airplane
control systems.
SOFT REAL TIME SYSTEMS
Deadline overruns are tolerable, but not desired.
There are no catastrophic consequences of
missing one or more deadlines.
Example: live video
streaming
EMBEDDED SYSTEMS
An embedded system is a computer system with
a dedicated function within a larger mechanical
or electrical system, often with a real time
constraint.
NEED OF ENERGY POWER EFFICIENCY
Critical design issue in real-time systems,
especially in battery- operated systems.
Smaller and lighter devices
Less heat generation
Better performance with unprecedented speed
Reduce cost
Reduce energy use
Longer battery life.
POWER ISSUE CAN BE ADDRESSED IN
Architecture Level
System Level
Application Level
SCHEDULING
Scheduling is method by which threads , process,
or data flow are given access to system resources.
This is usually done to load balance and share
system resources effectively.
There are various energy efficient scheduling
algorithms.
DYNAMIC VOLTAGE SCALING
Dynamic voltage scaling is a power management technique in computer architecture.
In which the voltage used in a component is increased or decreased, depending upon circumstances.
Dynamic voltage scaling to decrease voltage is known as undervolting. It is done in order to conserve power.
Dynamic voltage scaling to increase voltage is known as overvolting. It is done in order to increase computer performance.
SYSTEM MODELS
The real time system we are interested in consist of N independent periodic tasks, T = {τ1, τ2,⋅⋅⋅, τN} scheduled according to Earliest Deadline First (EDF).
τi = (Ci , Di ,Pi), is characterized by its worst case execution time Ci, deadline Di, and period Pi. We
assume Di<=Pi.
The Jth job of task τi is represented with Ji j = (ri j, ci j,di j),where ri j,ci j, and di j are the arrival time, actual execution cycles, and absolute deadline, respectively.
Each task τi is associated with a subset of
peripheral devices Φi = {M0,M1, ...,Mk}.
Φi can be further divided into shared part
Φi share and non-shared-part Φi nonshare.
Each peripheral device can be in one of the two
states Active State or Shut down state.
Energy overhead (Eo) andtime overhead (To) need
to be consumed to shut-down and later wake up
the processor or devices
Break even time is used to reflect
whether it is worthwhile to shut down the
processor /(peripheral devices) during idle
interval or not.
Ibe=max{Eo/P(idle),To}
DYNAMIC SCHEDULING ALGORITHM
Energy beneficial break even time Ieb is
computed dynamically for each job according to
the run-time conditions and shut down the
processor/devices whenever necessary.
THEOREM :
Given task set T = {τ1, τ2, ..., τN} with tasks
ordered by increasing value of Di.
All task deadlines can be guaranteed if any job
Ji of task τi is procrastinated by no more than Δi time units,
where Δi (called the procrastination time of task τi)
COROLLARY 1:
T – task set scheduled according to EDF
up(Ji)- upcoming job set in which each job Jk has
arrival time rk > ri .
If Ji is the only job in the ready queue, all jobs in
T can meet their deadlines if the starting
execution time of up(Ji) is delayed to tLS, where
tLS = min (rk +Δk) (1) Jk∈up(Ji )
COROLLARY 2:
T – task set scheduled according to EDF.
Ji current job to be executed in any time t. hp(Ji)
is the upcoming job set in which each job Jl has
arrival time rl > t and deadline dl < di.
Then all jobs in T can meet their deadlines if the
starting execution time of hp(Ji) is delayed to
˜tLS, where
˜tLS = min (rl +Δl) (2) l∈hp(Ji)
SCHEDULING ALGORITHM
1: Input: The current job Ji and the current
time t(cur);
2: if Ji is the only job in the ready queue then
3:Let edi = min{di, tLS};
//tLS is computed with 1
4: Compute Ii(eb) based on Ji’s current feasible
interval [t(cur),edi];
5: let fi be the expected completion time of Ji under
S(crit) ;
6: if (edi − fi) ≥ Ii(eb) then
CONT..
7: Execute Ji with S(crit) and shut down the
processor and devices in Φi at fi and set up the
wake up timer to be(tLS − fi);
8: // critical speed strategy with shut-down
9: else
10: Let t(na) be the earliest arrival time for the
upcoming jobs;
11: if fi ≥ t(na) and fi ≤ di then
12: Execute Ji with S(crit) ; // critical speed strategy
without shutting down shared devices in Φishare
13: else
14: Execute Ji with s′i = (ci×si)/(edi−(tcur)) non-
preemptively within [t(cur),edi]; // DVS strategy
stretching to edi;
µP
roc.
Sp
eed
Time
Start Deadline Start Deadline
Idle time
represent
s wasted
energy
Lower speed,
Lower voltage,
Lower energy
Energy ~ Work • Speed
Work
Work
CONT..
15: end if
16: end if
17: else
18: Compute ˜tLS based on equation (2);
19: Execute Ji with max{si, S(crit)} non-
preemptively within [t(cur),˜tLS] and shut down
its non-shared devices Φi nonshare upon its
completion;
20: end if
µP
roc.
Sp
eed
Time
S1 S2 S3 D2 D3
D1
W1 W2 W3
Task runs faster
to meet timing
constraints
APPLICATION AREAS Customer products:
Dish washers
Microwave ovens
Cars:
Airbag system
Engine control
Planes
Military
Weapons
Satellites
Robotics
Protection & security system
Intruder Alarm
Smoke/gas detection
LIMITATIONS
In DVS systems however, the performance level
is reduced during periods of low utilization such
that the processor finishes each task “just in time,” stretching each task to its deadline,
The primary caveat of overvolting is increased
heat: the power dissipated by a circuit increases
with the square of the voltage applied, so even
small voltage increases significantly affect power.
CONCLUSION
Dynamic Scheduling Algorithm is used to
minimize the system level energy consumption.
Approach of energy beneficial break even time is
is used to enhance the computation.
This approach can reduce energy consumption for
processor/devices significantly when compared
with other approaches.
REFERENCES
http://www.cse.unsw.edu.au/~cs9242/08/lectures/09-realtimex2.pdf - last seen 21/9/14
http://airccse.org/journal/ijcsea/papers/2212ijcsea16.pdf - last seen 24/9/14
http://www.dateconference.com/proceedings/PAPERS/2011/DATE11/PDFFILES/IP1_08.PDF - last
seen 21/9/14
http://www.ecs.umass.edu/ece/koren/architecture/RtDVS/intro.htm - last seen 28/9/14.