Академический Документы
Профессиональный Документы
Культура Документы
Sensor Networks
Wei Dong, Xue Liu , Chun Chen, Yuan He , Gong Chen, Yunhao Liu , and Jiajun Bu
Zhejiang Key Lab. of Service Robot, College of Comp. Sci., Zhejiang University
School of Comp. Sci., McGill University
Dept. of Comp. Sci. & Eng., Hong Kong University of Science and Technology
{dongw, chenc, chengong, bjj}@zju.edu.cn, xueliu@cs.mcgill.ca, {heyuan, liu}@cse.ust.hk
I. I NTRODUCTION
A fundamental challenge in wireless networks is that radio
links are subject to transmission power, fading, and interference, which degrade the data delivery performance. This
challenge is exacerbated in wireless sensor networks (WSNs),
where severe energy and resource constraints preclude the
use of many sophisticated techniques that may be found in
other wireless systems [1]. For example, (i) bit rate adaptation
protocols [2] demand special hardware that is not available on
general sensor nodes; (ii) effective forward error correction
(FEC) requires the amount of redundant data transmitted to
be tuned to match the link qualities [3], which is difcult
to achieve in dynamic WSNs; (iii) other sophisticated coding
schemes [4], [5] incur delays and computation overheads that
are only suitable for specic applications.
In this paper, we consider a simple, cost-effective solution
based on the technique of dynamic packet length control to
improve the performance in these varying conditions. A tradeoff exists between the desire to reduce header overhead by
making packet large, and the need to reduce packet error rates
(PER) in the noisy channel by using small packet length [6].
Although there have been several studies on packet length
optimizations in the literature [6][11], existing approaches
are not practically applicable to the sensor network scenario.
A packet optimization scheme applicable in sensor networks
must have the following important features.
II. M OTIVATION
To better understand how a practical packet length adaptation scheme should be designed, we ran a series of experiments
to provide insights that guide our design.
A. Packet Length Optimization
To see the impact of variable packet length to the system
performance in WSNs, we setup two TelosB motes which
communicate variable packets at a distance of approximately
8 meters with transmission power level 3 in a quiet indoor
environment. We measure the packet reception rate (PRR)
and the transmission efciency (E, which is dened as the
received useful bytes divided by the overall transmitted bytes)
for each packet payload length. A high E value indicates a
high goodput and a high energy efciency provided that the
transmission time and the transmission energy consumption
are approximately linear to the packet length. We ran each
experiment 30 times, and Figure 1 shows the mean value
and the standard deviation of PRR and E. We can see that
with the packet payload length increases, PRR decreases while
E reaches its maximum value at packet payload length 30
(bytes). Using the optimal packet length improves approximately 40% performance in terms of E compared to using the
smallest packet length in this case.
We have also run other experiments at different distances
and transmission powers. Results show that the optimal packet
length varies in different conditions. Therefore, packet length
optimization is benecial in WSNs.
0.50
0.8
0.45
0.7
0.40
0.6
0.35
0.5
0.30
0.4
0.25
10
20
30
40
50
60
70
80
PRR
0.3
90
-50
-60
RSSI (dbm)
-70
-80
-90
-100
3000
6000
9000
12000
15000
Time (ms)
1.0
0.9
PRR
0.8
0.7
0.6
0.5
20
30
-84
40
50
60
Time (seconds)
70
80
90
100
RSSI
-86
-88
-90
-92
110
100
LQI
90
80
70
60
50
40
directly measured RSSI and LQI are not enough for accurately
estimating link qualities.
It is also worth mentioning that link estimation using proactive beaconing is also inappropriate in our scenario, because
the data packet length is different from that of beacons, and
it varies with link conditions. Therefore, data-plane PRR
statistics are needed in order to improve the accuracy of link
quality estimations.
III. D ESIGN
In this section, we present the design of DPLC, a dynamic
packet length control scheme for WSNs. Below, we identify
the major design goals.
Dynamic adaptation. DPLC should provide a dynamic
adaptation scheme to achieve performance improvements
in dynamic, time-varying sensor networks.
Accurate link estimation. DPLC should implement an
accurate link estimation method that can capture both
physical channel conditions (due to channel fading, mobility, or power degradation) and interferences (from
exposed and hidden terminals).
Ease of programming. DPLC should provide easy-touse services to facilitate upper-layer application programming.
Lightweight for implementation. DPLC should be
lightweight for resource constrained sensor nodes.
We will see how DPLC achieves the rst two design
goals, i.e., dynamic adaptation and accurate link estimation
in Section III-C. The remaining two design goals, i.e., ease
of programming and lightweight for implementation, will be
elaborated in Section IV-A and Section IV-B respectively.
A. Overview
The DPLC scheme works as follows. The application passes
an application-level message for transmission. The DPLC
module at the sender decides whether to use the aggregation
service (AS, if the message length is small) or the fragmentation service (FS, if the message length is larger than the
maximum packet length supported by the radio, i.e., 128 bytes
for CC2420). The link estimator within DPLC dynamically
estimates the optimal packet length for transmission. Based
on this, the DPLC module at the sender decides how many
messages should be aggregated (for AS), or how many frames
the message should be fragmented into (for FS). When a
frame is ready for transmission (enough messages have been
aggregated or time is out in AS), we actually send it via the
MAC layer. When the DPLC module at the receiver receives
a MAC frame, it deaggregates or defragments the frame in
order to obtain the original message. When the message is
ready (all frames in the message have received or the receive
buffer is full in FS), the DPLC module at the receiver noties
the upper layer for further handling.
As mentioned above, the DPLC scheme provides two services for upper-layer applications, i.e., the aggregation service
(AS, for small messages) and the fragmentation service (FS,
for large messages).
(i) AS is useful for small data collection, e.g., CTP [16]. AS
provides three distinct mechanisms, i.e., reliable transmissions
(AS ), unreliable transmissions with xed number of retransmissions (ASn , where n 1 is the retransmission number), and
unreliable transmissions (AS0 ). Both AS and ASn requires
L2 ACKs provided by the link layer, because packets need to
be retransmitted (at least once) when they are lost. For AS0 ,
we additionally provide a more efcient ACK scheme called
AggAck that does not rely on L2 ACKs, and thus mitigate the
ACK overhead (we use AS0 -L2 to denote AS0 with L2 ACKs
and AS0 -AA to represent the one with AggAck afterwards).
(ii) FS is useful for bulk data transmission, e.g., Flush
[18]. FS provides reliable transmissions as a large message is
usually very important for upper-layer applications. FS does
not necessarily depend on L2 ACKs. As mentioned above, we
additionally provide the AggAck mechanism to mitigate the
ACK overhead, and more importantly, to deal with data packet
retransmissions (we use FS-L2 to denote FS with L2 ACKs
and FS-AA to represent the one with AggAck afterwards).
The link estimation method within DPLC dynamically estimates the link quality based on passive observations of packet
receptions [19]. For each outgoing link, the sender keeps a
sliding window in which each bit tracks whether the sent
packet was ACKed. Based on this, it tunes the packet length
for a given link. The link estimation method does not necessarily depend on L2 ACKs. As mentioned above, we additionally
provide AggAck to provide such information. The advantages
of AggAck are three-fold. First, for AS0 and FS, it mitigates
L2 ACK overhead for each transmitted packet. Second, for FS,
it reduces redundant data packet retransmission in asymmetric
links (in which the reverse link quality for ACK is poor).
Finally, in the presence of asymmetric links, it provides an
accurate estimation of directional links which is important to
optimize multi-hop data delivery performance of a single ow.
1
k1 (l) + k1
i=1 i
dri = 1 (1 pi )m+1
(4)
(3)
interface DMSend {
command error_t send(am_addr_t addr,
void* msg,
uint16_t len);
event void sendDone(void* msg, error_t err);
command void* getPayload();
command void setLinkAck(bool ack);
command void setRetries(uint8_t num);
command void flush(am_addr_t addr);
command void setMaxLen(uint8_t len);
command void setMinLen(uint8_t len);
}
(a) The DMSend interface
interface DMReceive {
event void* receive(void* msg, uint8_t len,
uint16_t pktlen);
command message_t* getAM(message_t* msg);
}
(b) The DMReceive interface
Typ Ack
Typ
Count
Offset
Offset (cont.)
TE (E)
1 byte
1 byte
BER=8E-4
22
12m
2
18
19
20
(a)
BER=1E-4
16
BER=8E-4
17
3
14
10m
12
0.6
15
22
(b)
13
10
11
0.5
0.4
0.3
0.2
0.1
0.0
Fix-10
Fix-110
Fix-40 (opt)
AS0-L2
AS0-AA
Schemes
0.30
B. Testbed Results
0.25
0.20
0.15
0.10
0.05
0.00
(in which the ACK overhead is not accounted for), and the
transmission efciency with ACK overhead (in which the ACK
overhead is also accounted for). We can see that in both
cases, DPLC achieves more than 90% performance compared
to the optimal scheme. We can also see that the AggAck
mechanism mitigates the ACK overhead, leading to about 5%
improvement considering the ACK overhead.
Fix-10
Fix-110
Fix-110,30 (opt)
AS0-L2
AS0-AA
Schemes
1.00
0.20
CTP
CTP-DPLC
CTP-max
0.90
0.85
0.80
CTP
CTP-DPLC
CTP-max
0
60000
0.18
0.16
0.14
55000
50000
Duty cycle
Delivery rate
0.95
65000
45000
0.08
0.06
35000
0.04
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
CTP
CTP-DPLC
CTP-max
0.02
30000
0
10
20
30
40
50
60
70
0.00
80
10
Time (min)
20
30
40
50
60
70
80
Time (min)
1.0
60000
0.18
0.9
0.7
0.6
0.5
0.4
0.3
0.2
0.16
50000
0.14
40000
0.12
Duty cycle
0.8
Delivery rate
0.10
40000
Time (min)
30000
20000
0.10
0.08
0.06
0.04
10000
0.02
0.1
0.0
0.12
CTP
CTP-DPLC
CTP-max
Schemes
CTP
CTP-DPLC
CTP-max
Schemes
0.00
CTP
CTP-DPLC
CTP-max
Schemes
[3] J. W. Hui and D. Culler, The Dynamic Behavior of a Data Dissemination Protocol for Network Programming at Scale, in Proceedings of
ACM SenSys, 2004.
[4] M. Yang and Y. Yang, A Hypergraph Approach to Linear Network
Coding in Multicast Networks, IEEE Transactions on Parallel and
Distributed Systems, 2009, to appear.
[5] Y. Yu, B. Krishnamachari, and V. K. Prasanna, Data Gathering with
Tunable Compression in Sensor Networks, IEEE Transactions on
Parallel and Distributed Systems, vol. 19, no. 2, pp. 276287, 2009.
[6] P. Lettierri and M. B. Srivastava, Adaptive Frame Length Control for
Improving Wireless Link Throughput, Range, and Energy Efciency,
in Proceedings of IEEE INFOCOM, 1998.
[7] C. K. Siew and D. J. Goodman, Packet Data Transmission Over Mobile
Radio Channels, IEEE Transactions on Vehicular Technology, vol. 38,
no. 2, pp. 95101, 1989.
[8] E. Modiano, An adaptive algorithm for optimizing the packet size used
in wireless ARQ protocols, Wireless Networks, vol. 5, no. 4, pp. 279
286, 1999.
[9] P. R. Jelenkovic and J. Tan, Dynamic Packet Fragmentation for Wireless
Channels with Failures, in Proceedings of ACM MobiHoc, 2008.
[10] Y. Sankarasubramaniam, I. F. Akyildiz, and S. W. Mclaughlin, Energy
Efciency based Packet Size Optimization in Wireless Sensor Networks, in Proceedings of IEEE Internal Workshop on Sensor Network
Protocols and Applications, 2003.
[11] M. C. Vuran and I. F. Akyildiz, Cross-layer Packet Size Optimization
for Wireless Terrestrial, Underwater, and Underground Sensor Networks, in Proceedings of IEEE INFOCOM, 2008.
[12] GreenOrbs, Available: http://www.greenorbs.org.
[13] L. Mo, Y. He, Y. Liu, J. Zhao, S. Tang, X.-Y. Li, and G. Dai, Canopy
Closure Estimates with GreenOrbs: Sustainable Sensing in the Forest,
in Proceedings of ACM SenSys, 2009.
[14] TinyOS, Available: http://www.tinyos.net.
[15] P. Levis, N. Lee, M. Welsh, and D. Culler, TOSSIM: Accurate and
Scalable Simulation of Entire TinyOS Applications, in SenSys, 2003.
[16] O. Gnawali, R. Fonseca, K. Jamieson, D. Moss, and P. Levis, Collection
Tree Protocol, in Proceedings of ACM SenSys, 2009.
[17] K. Srinivasan and P. Levis, RSSI is Under Appreciated, in Proceedings
of EmNets, 2006.
[18] S. Kim, R. Fonseca, P. Dutta, A. Tavakoli, D. Culler, P. Levis, S. Shenker,
and I. Stoica, Flush: A Reliable Bulk Transport Protocol for Multihop
Wireless Networks, in Proceedings of ACM SenSys, 2007.
[19] G. Hackmann, O. Chipara, and C. Lu, Robust Topology Control for
Indoor Wireless Sensor Networks, in Proceedings of ACM SenSys,
2008.
[20] L. Sang, A. Arora, and H. Zhang, On Exploiting Asymmetric Wireless
Links via One-way Estimation, in Proceedings of ACM MobiHoc, 2007.
[21] J. D. Spragins, J. L. Hammond, and K. Pawlikowski, Telecommunications: Protocols and Design. Addison Wesley Publishing Company,
1991.
[22] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, Wireless
sensor networks: A survey, Computer Networks, vol. 38, pp. 393422,
2002.
[23] X. Liu, Q. Wang, W. He, M. Caccamo, and L. Sha, Optimal realtime sampling rate assignment for wireless sensor networks, ACM
Transactions on Sensor Networks, vol. 2, no. 2, pp. 263295, 2006.
[24] S. Ganeriwal, I. Tsigkogiannis, H. Shim, V. Tsiatsis, M. B. Srivastava,
and D. Ganesan, Estimating clock uncertainty for efcient duty-cycling
in sensor networks, IEEE/ACM Transactions on Networking, vol. 17,
no. 3, pp. 843856, 2009.
[25] H. Zhang, A. Arora, and P. Sinha, Learn on the Fly: Data-driven Link
Estimation and Routing in Sensor Network Backbones, in Proceedings
of IEEE INFOCOM, 2006.
[26] T. He, B. M. Blum, J. A. Stankovic, and T. Abdelzaher, AIDA:
Adaptive Application-Independent Data Aggregation in Wireless Sensor
Networks, ACM Transactions on Embedded Computing Systems, vol. 3,
no. 2, pp. 426457, 2004.