CprE 458/558: Real-Time Systems

1
CprE 458/558 Real-Time Systems

• Energy-aware QoS packet scheduling

2
Overview

• Motivation
• Key hardware techniques for energy savings
• Energy-aware weighted fair queuing
• Energy-aware real-time packet scheduling

3
Motivation
Energy Consumption
QoS / Real-time guarantees
4
Key hardware techniques

• Dynamic Voltage Scaling (DVS) for processor
  energy savings
• Dynamically vary the operating voltage
  frequency of the processor to reduce energy
  consumption
• Dynamic Modulation Scaling (DMS) for wireless
  radio energy savings

5
Dynamic Voltage Scaling

• Energy consumption of task with cc number of
  computation cycles operated at a voltage V and a
  corresponding frequency f is given by
• E CC V2 CC F2
• Time taken to complete the task is given by
• T CC / F
• Therefore we can run a task at a lower frequency
  and reduce energy consumption. However, you will
  need relatively more time to complete the task.

6
Dynamic Modulation Scaling (DMS)
The energy consumption of the radio in
transmitting a bit at a modulation level b is
given by

The transmission time a bit at a modulation level
b (number of bits per symbol) is given by
Where Rs is the number of symbols sent over the
channel per sec.
7
DMS Energy-Delay tradeoffs
8
Problem - 1

• To assign modulation levels to the incoming
  traffic flows while guaranteeing delay bounds
  within the WFQ framework.

9
E2WFQ scheduler
10
The WFQ scheduler bounds

• Traffic flow model Leaky bucket regulated flow
• If a flow Ai (si, ?i) is guaranteed a rate of gi,
  then the maximum delay Di under GPS is given by
• Di si / gi
• The maximum delay Di under WFQ is given by
• Di si / gi Lmax / C
• where Lmax is the maximum packet size
• Where C is the link capacity

11
Important Observation

• ?i , the input rate of an input stream is much
  lower than its guaranteed rate gi
• Therefore, operating at the link transmission at
  the instantaneous rate will result in energy
  savings

12
E2WFQ scheduler basic idea

• Monitor the instantaneous input rate
• Adapt the transmission rate to the input rate
  subject to the delay constraints

13
Monitoring the input rate

• Instantaneous queue size (number of packets) is a
  good indicator of the instantaneous input arrival
  rate
• If input rate is greater than the output rate the
  queue size increases
• On the other hand, if the input rate is lesser
  than the output rate the queue size decreases
• This where we can apply DMS to reduce energy
  consumption

14
A typical inflow rate profile
Peak rate
Rate
Guaranteed rate (gi)
Average rate (?i)
time
15
Delay constraints

• Let ? be the desired time from the packets
  arrival at the end of the queue to its departure
  from the head of the queue
• Let there be m packets (P1, P2 Pm ) in the
  queue arrived at times (A1, A2 Am )
  respectively. Let, Am T current time and
  further assume each packet of low i is of size
  Li

Pm
Pk
P1
K Li

• What should be the output rate ( ri,k ) of the
  flow i to guarantee the ? delay constraint to a
  packet Pk ?

16
The instantaneous output rate
The output rate Rout,i for a particular flow i
Guaranteed rate
Maximum of the output rates required by all the
queued packets
The total output rate of the link
17
The instantaneous modulation level
The modulation level for the link with a capacity
C is given by
18
Maximum delay expressions
Theorem The maximum packet of delay of stream
i , under the E2WFQ scheme is given by
19
Energy aware real-time packet scheduling
Sensor nodes send real-time (periodic) multimedia
streams to the aggregation node G.
20
Problem

• Assign modulation levels to each of the packets
  to reduce energy consumption subject to the
  real-time deadline constraints.
• This is very similar to the DVS scheduling of
  periodic tasks at the processor.
• Unlike the tasks on a processor, the messages on
  the communication link cannot be pre-empted.

21
The Real-Time DMS packet Scheduler
Admission Controller
RT-DMS Scheduler
Periodic RT Messages
22
Admission test
The following time completion test is employed
for admission of periodic streams
23
Static DMS

• Assuming maximum packet size for all the admitted
  packets, find the least modulation level which
  ensures all the deadlines
• This can be accomplished an iterative approach
  trying each modulation level for all the packets

24
Dynamic DMS

• The packet sizes exhibit variations, the exact
  packet size is known before the transmission.
• The idea behind dynamic DMS is to reduce the
  modulation level of a smaller packet so that it
  takes as much time as the maximum sized packet
  would have taken

25
Stretch DMS

• If the finish time of the current packet
  transmission and the arrival time of the next
  packet transmission are unequal. Some amount of
  slack will be left unused or the link will idling
  during that slack.
• We can further reduce the modulation level to
  exploit the entire slack.

26
RT-DMS example with 3 periodic streams
No DMS
Static DMS
Run-time
Dynamic DMS
Stretch DMS
27
Some References

• 1 V. Raghunathan et al, E2WFQ An energy
  efficient fair scheduling policy for wireless
  systems, ISPLED 2002.
• 2 C. Schurgers et al., Modulation scaling for
  real-time energy aware packet scheduling,
  GLOBECOM 2001.

28
Thank You!!