A PowerAware and QoSAware Service Model on Wireless Networks

1
A Power-Aware and QoS-Aware Service Model on
Wireless Networks

• Hao Zhu and Guohong Cao
• Department of Computer Science Engineering
• The Pennsylvania State University
• Infocom 2004

Present Yi-Wei Ting
2
Outline

• Introduction
• Related Works
• Proposed PBS (priority based bulk scheduling)
  model
• Performance Evaluation
• Conclusion
• Comment

3
Introduction

• Wireless network interface (WNI) accounts for
  power consumed by mobile terminals (MTs).
• Putting the WNI into sleep when it is idle is an
  effective technique to save power.
• How to achieve power saving without violating QoS
  is a challenge for supporting streaming
  applications on wireless networks.

4
Related Works

• IEEE 802.11 supports power saving mode in which
  the WNI only needs to be active periodically.
• 1. The WNI in sleep mode only wakes up
  periodically to check for possible incoming
  packets from the BS.
• 2. The BS transmits a beacon frame after a
  regular beacon interval.
• 3. In each beacon frame, a traffic indication map
  contains information about which WNI has buffered
  incoming packets.
• 4. If the WNI finds that it has incoming packets,
  it should stay active to receive the packets.

5
Scheduling algorithm

• A scheduling algorithm can be classified as two
  method.
• Work-conserving scheduling
• A server is never idle when there is a packet to
  send.
• Non-work-conserving scheduling
• A packet is not served until it is eligible, even
  though the server is idle at that time.

6
Work-conserving scheduling

• A server is never idle when there is a packet to
  send.
• disadvantage MT does not know the following
  service sequence due to lack of global
  informatioin regarding the scheduling pattern of
  all flows in the system.

MT1 act_time5B/C, rec_timeB/C, 80
of the power wasted (B send B bits,
C capacity)
0.2C
0.3C
0.5C
7
Non-work-conserving scheduling (NVC)

• Let the WNI enter sleep when it is not used
• BS and MT mutaully agree on a scheduling pattern,
  when the BS send sends a packet to the MT, the
  scheduler piggybacks the information about the
  eligible time the next packet to be transmitted.
• The eligible time can be calculated based on the
  flows data rate.

Only Ton?off Toff?on ltlt Tintval can save power,
Ton?off10ms (Ref.)
8
Bulk Scheduling (BKS)

• The channel is divided into bulk slots, each slot
  equal to B/C.
• Bulk scheduling cannot provide QoS when multiple
  flows request data at the same time.
• Example if at t1, suppose MT3 and MT1 will miss
  their deadline if waiting for another B/C time
  slot, scheduling MT1 to serve will force MT3 to
  miss its deadline.

9
Proposed PBS model

• The basic idea of PBS is to let the MT buffer as
  much data as possible without affecting the QoS
  requirement of other flows.
• Relying on the buffered data,the MT can put its
  WNI into sleep and wake up only when the
  prefetched data is not enough to satisfy its QoS
  requirement.

10
Proposed PBS model

• A scheduler at the BS side
• Control the channel access among multiple MTs.
• A proxy at the MT side
• Coordinate with the scheduler at the MT side.

11
1) The PBS Scheduler

• Each packet of a flow is assigned a deadline.
• The scheduler orders the transmission of packets
  according to their deadlines.
• If a flows aggregated service goes beyond the
  minimum service required to maintain the QoS.
• it will be removed from the scheduling region
  until it needs more data to maintain the QoS.

12
Scheduling state management

• For the purpose of flow control, there is an
  upper limit of ahead-service for each flow fi,
  denoted by Maxservi.
• When aheadi gt Maxservi, the scheduler stops
  serving it and changes statei to idle.
• Suppose the scheduler provides enough
  ahead-service to fi i.e., aheadi gt Ø, (Ø is a
  system parameter to represent the lower bound for
  ahead-service).

13
Example 3 flows, data rate1kbps, C10kbps,
packet length1kbytes
14
Example WFQ scheduling
15

• Notations
• A the set of flows in active state, t the
  current time
• Di the deadline of the head-of-line packet of fi
• Schedule()
• 1 begin
• 2 if (ANULL)
• 3 idle in the time slot goto
  begin
• 4 if (no primary flow)
• 5 select the primary flow fi according
  to Eq.(3)
• 6 if (t gt arg minDj and j ! i)
• 7 i j // the deadline of
  the secondary flow fj will be violated, so serve
  fj
• 8 p fi.deque()
• 9 If (aheadi gt Maxservi or (aheadi gt Ø and
  j(fj ? A and aheadj lt Ø)
• 10 mark(p)
• 11 send(p)
• 12 if (the transmission is successful)
• 13 Di Di p.length / ri
• 14 if (p is marked)

• fprim arg min

(Ø aheadi(t)) ri
(C - Sj ?A rj
i ?A
16
2) The PBS proxy

• Calculate the ahead-service of each flow
  according to the flow according to the flows
  data rate, packet length and the arrival timeo of
  each packet.
• If the proxy finds that the packet is marked, the
  WNI will be shutdown for a time period equal to
  the calculated ahead-service.

17
Performance Evaluations

• Simulation time 100 seconds.
• Audio-on-Demand (AoD)
• Demo MP3, interval is distributed randomly
  between 0.0, 2.0.
• Maxserv 2000ms for AoD flows
• WWW service
• ON/OFF Traffic model
• ON period mean of 12KB.
• OFF period mean of 2 seconds.
• Ø 80ms
• Wireless channel 384Kbps.
• Time slot 2.5ms
• Each flow 56Kbps
• Ton?off5ms, Toff?on10ms.

18
Performance metrics

• The QoS (NIT Throughput)
• The time spent in active, sleep and state
  transition
• Noticeable interrupt time (NIT)
• Only record when interrupts are greater than 20ms
  
• Amount of prefetched data
• Buffer trace of AoD flow

19
The BKS scheduler only randomly selects one
winning flow to serve
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Conclusion

• Proposed a deadline-based priority bulk
  scheduling (PBS) service model to save power and
  QoS quarantees.