Siphon at SenSys05

Views:
 
Category: Education
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Siphon: Overload Traffic Management using Multi-Radio Virtual Sinks in Sensor Networks: 

Siphon: Overload Traffic Management using Multi-Radio Virtual Sinks in Sensor Networks Chieh-Yih Wan, Intel Research Shane B. Eisenman, Columbia University Andrew T. Campbell, Dartmouth College Jon Crowcroft, Cambridge University

The Problem: 

The Problem Observations Funneling Effect limits performance Congestion Collapse Existing congestion control techniques are effective at reducing packet loss, but do little to help data fidelity Result We live in a fidelity-limited world Broader Challenge How do we increase fidelity in sensor networks Siphon’s Contribution To offer increased fidelity during periods of congestion and traffic overload in sensor networks

Funneling Effect: 

Funneling Effect Many-to-one traffic pattern causes congestion in the routing funnel

The Problem: 

The Problem Observations Funneling Effect limits performance Congestion Collapse Existing congestion control techniques are effective at reducing packet loss, but do little to help data fidelity Result We live in a fidelity-limited world Broader Challenge How do we increase fidelity in sensor networks Siphon’s Contribution To offer increased fidelity during periods of congestion and traffic overload in sensor networks

Congestion Collapse: 

Congestion Collapse * From “Mitigating Congestion in Wireless Sensor Networks”, SenSys’04. Results from a 55 node Mica2 indoor testbed (office environment)

The Problem: 

The Problem Observations Funneling Effect limits performance Congestion Collapse Existing congestion control techniques are effective at reducing packet loss, but do little to help data fidelity Result We live in a fidelity-limited world Broader Challenge How do we increase fidelity in sensor networks Siphon’s Contribution To offer increased fidelity during periods of congestion and traffic overload in sensor networks

Existing Congestion Control Techniques: 

Existing Congestion Control Techniques Fusion, CODA, ESRT use rate control and packet drop techniques to control congestion * From results presented in “CODA: Congestion Detection and Avoidance in Sensor Networks”, SenSys’03

The Problem: 

The Problem Observations Funneling Effect limits performance Congestion Collapse Existing congestion control techniques are effective at reducing packet loss, but do little to help data fidelity Result We live in a fidelity-limited world Broader Challenge How do we increase fidelity in sensor networks Siphon’s Contribution To offer increased fidelity during periods of congestion and traffic overload in sensor networks

Siphon: 

Siphon Add capacity on-demand by deploying a multi-radio overlay mesh based on “virtual sinks”

Virtual Sink Discovery: 

Virtual Sink Discovery Physical Sink Virtual Sink Mote VS Neighbor Default Route

Traffic Redirection: 

Traffic Redirection

Design Considerations: 

Design Considerations Virtual Sink placement Advertisement scope Placement density Guidelines on when to redirect traffic to the Virtual Sink

Virtual Sink Advertisement Scope: 

Virtual Sink Advertisement Scope Simulation w/ 30 nodes 1 Virtual Sink Several randomized topologies

Virtual Sink Deployment Density: 

Virtual Sink Deployment Density

Traffic Redirection Guidelines: 

Traffic Redirection Guidelines Redirect to Virtual Sinks only when local congestion is inferred, via channel load estimate or buffer occupancy threshold. Virtual sink neighbor must have link quality < A% worse than that of the default next hop to avoid forcing the use of lossy links. To avoid the possibility of routing loops, Virtual Sink neighbors that are upstream from the congested node are not used.

TestBed Details: 

TestBed Details 48 Mica2 motes in a 6x8 multi-hop grid (grid calibration: 1-hop  >80%, 2-hop  <20%) Stargate platform with IEEE 802.11b and Mica2 TinyOS-1.1.0 (Surge, MultiHopRouter) Performance Intuition: Energy Tax and Fidelity performance should improve with increasing load and with the addition of Virtual Sinks.

Uniform Packet Generation (where 48 nodes are srcs): 

Uniform Packet Generation (where 48 nodes are srcs) Virtual sinks increase fidelity and energy tax savings 55% Fid. Boost 25% Fid. Boost 45% Tax Reduction 25% Tax Reduction 10% 5% 1 2

Sparse Packet Generation (where 3 nodes are srcs): 

Sparse Packet Generation (where 3 nodes are srcs) Siphon provides improved performance versus rate-limit/pkt drop techniques Generic data dissemination app. Results avg. 5 arbitrary placements of 1 Virtual Sink 20% Fidelity Boost 2x reduction in pkt loss

Energy Usage: 

Virtual Sink Usage Cost = Energy w/ VS Energy w/o VS Energy Usage Using Virtual sinks reduces the cost of delivering packets to the Physical Sink

Load Balancing: 

Load Balancing Residual Energy = Remaining Energy Initial Energy Placing Virtual Sinks spreads the traffic load more equally NS2 Simulation 70 nodes uniformly dist’d 3 Virtual Sinks randomly 1/3 VS is the Physical Sink Complementary CDF shows the probability a given node has a residual energy higher than X%

Related Work: 

Related Work First-generation congestion control algorithms Hierarchical Sensor Networks XScale (expedited Delivery) Tenet (hierarchy for scalability) Intel Clustering algorithms in MANET

Conclusion: 

Conclusion Contribution Boosts Fidelity to the application during periods of traffic overload Provides a positive Energy Tax Savings in the face of network congestion. Interoperates with existing congestion control schemes (e.g. CODA) Siphon algorithms more generally apply to heterogeneous/hierarchical sensor networks (storage, aggr, comp.)

Thanks for listening.: 

Thanks for listening. Contact: shane@ee.columbia.edu

authorStream Live Help