280 likes | 451 Views
Sprinkler: A Reliable and Energy Efficient Data Dissemination Service in Extreme Scale Wireless Networks of Embedded Devices. Vinayak Naik , Anish Arora, Prasun Sinha, and Hongwei Zhang Dependable Distributed and Networked Systems December 7, 2005.
E N D
Sprinkler: A Reliable and Energy Efficient Data Dissemination Service in Extreme Scale Wireless Networks of Embedded Devices Vinayak Naik, Anish Arora, Prasun Sinha, and Hongwei Zhang Dependable Distributed and Networked Systems December 7, 2005
New Model due to Extreme Scale Wireless Embedded Devices • Embedded devices are constrained in following resources: • CPU • Memory • Power • Characteristics of wireless medium • Spatial and Temporal variation in link quality • Hidden terminal effect • Extreme scale demands sub-linear time complexity • O(n) isn’t good enough for resource constrained devices • Different model as compared to that of the Internet • Existing network services may not work
Outline • Motivation and Requirements • Insight behind Solution • Formal Problem Statement and Algorithms • Analysis and Comparison • Conclusion
Irrigating ExScal • Motivation behind data dissemination service • Reprogramming in the field (hundreds of packets) • System reconfiguration (tens of packets) • Health monitoring (< ten packets) • Problem of bulk data dissemination service • 100% Reliability • Energy efficiency • Low latency
Outline • Motivation and Requirements • Insight behind Solution • Formal Problem Statement and Algorithms • Analysis and Comparison • Conclusion
Energy Saved is Energy Generated • Load shedding • Packet Transmissions • Microprocessor and Idle Radio (Not covered in this talk)
Unit Disk Model R B R A R = Transmission Radius
CDS Connected Dominating Set Fewer number of senders
Collision! Lost packet Hidden Terminal Effect B C A
R R Time Division Multiple Access D C B A Schedule transmissions
Outline • Motivation and Requirements • Insight behind Solution • Formal Problem Statement and Algorithms • Analysis and Comparison • Conclusion
Formal Problem Statement • Divide-n-Conquer • An algorithm to compute a CDS, of size O(1) times the minimum, in O(1) time • An algorithm to compute a distance-2 vertex coloring, with O(1) times the minimum # of colors, in O(1) time • A reliable data dissemination protocol that utilizes a CDS and a corresponding distance-2 vertex coloring • Assumptions • Minimum density: ≥ 1 node per square of length • Location information
O(1) Algorithm to Compute CDS • Division of network into disjoint square-shaped clusters,each of length • Election of a cluster-head in each cluster • Decision whether a cluster-head belongs to CDS or not • Variables: • r be the total number of cluster-heads in X axis • c be the total number of cluster-heads in Y axis • u(i,j) be any cluster-head and (i,j) be its (X,Y) coordinates • Program: A node u(i,j) ∈ M, where 0 ≤ i ≤ r−1 and 0 ≤ j ≤ c−1, if • r mod 3 ≡ 0 : [i mod 3 ≡ 1] ∨ [(i mod 3 ≡ 1) ∧ (0 < i < r−1) ∧ (j = 0)] • r mod 3 ≡ 1 : [i mod 3 ≡ 0] ∨ [(i mod 3 ≡ 0) ∧ (j = 0)] • r mod 3 ≡ 2 : [i mod 3 ≡ 1] ∨ [(i mod 3 ≡ 1) ∧ (i ≡ 0) ∧ (j = 0)]
Performance Ratio = R Clustering, Selecting cluster-heads, Computing CDS CDS Computation
0 1 2 3 4 5 6 7 6 7 14 15 Performance Ratio = 14 15 0 1 2 3 4 5 6 7 D-2 Vertex Coloring < 2R R 8 9 10 11 12 13 R Numbers indicate colors.
Data Dissemination Protocol • Streaming phase • Only CDS nodes transmit • Transmissions in TDMA slots • Results in reliable data dissemination to all CDS nodes • Recovery phase • Any node can transmit • Unscheduled transmissions • Results in reliable data dissemination to all the nodes
E 2 0 0 1 1 0 1 0 2 1 3 1 1 1 4 2,1 2,1 Recovery Packet Lost! Empty Slot Empty Slot Recovery Req Recovery Req Streaming Phase A B C D R 2
Models for Real Radio • Radio models in real environment are more complex than unit disk model • Packet delivery rate for XSS in an outdoor environment • Similarly, for indoor testbeds
Adapting Sprinkler to Real Radio Models • Input parameter • Transmission radius ( ) • Procedure • Initialize = , where is the reliable communication range (100% packet delivery) • Keep incrementing till the number of transmissions for the test broadcast are reducing • Density assumption still holds • Since every square of length contains at least one node, every square of length also contains at least one node
Outline • Motivation and Requirements • Insight behind Solution • Formal Problem Statement and Algorithms • Analysis and Comparison • Conclusion
Anatomy of XSS • XSS: Extreme Scaling Stargate • Stargate • SMC 2532W-B High Power IEEE 802.11b PCMCIA card • BU-303 GPS mouse via USB • External antenna connection
Kansei [The 2nd TinyOS Technology Exchange at Berkeley, 2005] • A testbed containing 200 pairs of XSSs and XSMs • A multi-hop IEEE 802.11 network • Using attenuators and S/W Tx power control • Applications • Debugging • Measuring performances of protocols • Web interface for experimentations • http://exscal.nullcode.org/kansei
Scalability of Sprinkler Hops Density
Comparison • Existing reliable bulk data dissemination services • Deluge • Infuse • MNP • PSFQ • Deluge protocol • Doesn’t uses CDS and TDMA • Uses sender suppression technique to reduce number of packet transmissions • Commonly used service for mote reprogramming • Simulation and experiment setup • A 7x7 network with a base station at a corner • Payload of 240 packets
Performance: # Packet Transmissions Deluge Sprinkler Source Source
Performance: Latency Deluge Sprinkler Source Source
Outline • Motivation and Requirements • Insight behind Solution • Formal Problem Statement and Algorithms • Analysis and Comparison • Conclusion
Conclusion • Sprinkler: Reliable and energy efficient data dissemination service[The 26th IEEE Real-Time Systems Symposium at Miami, 2005] • Energy efficient • Reduces # packet transmissions • Scalable • Constant time algorithms • Low latency • Pipelines transmissions in space • Future work • Use of hexagon-shaped clusters instead of square-shaped clusters • CDS and D-2 vertex coloring in the presence of holes of bounded size and regular shape