Topology Formation and Public Policy

Topology Formation
and
Public Policy
Jeff Pang
15-848E
Topology Formation Overview
• Principles and Protocols for Power Control in Ad Hoc Networks,
V. Kawadia and P. R. Kumar, IEEE Journal on Selected Areas in
Communications.
• Vikas Kawadia and P. R. Kumar, ``A Cautionary Perspective on
Cross Layer Design.'' To appear in IEEE Wireless
Communication Magazine.
• Roger Wattenhofer, Li Erran Li, Victor Bahl and Yi-Min Wang,
Distributed Topology Control for Power Efficient Operation in
Multihop Wireless Ad Hoc Networks. Proc. of IEEE INFOCOM,
pages 1388-1397, April 2001
• Ning Li and Jennifer C. Hou, Topology control in heterogeneous
wireless networks: problems and solutions, in Proc. of IEEE
INFOCOM 2004, March, 2004.
Power Control
1mW
4 mW
1 mW
4mW
Power Control:
Cross-Layer Design Issues
• Physical Layer
– Power control affects quality of signal
• Link Layer
– Power control affects number of clients sharing channel
• Network Layer
– Power control affects topology/routing
• Transport Layer
– Power control changes interference, which causes
congestion
• Application/OS Layer
– Power control affects energy consumption
Cross Layer Design Example:
Rate Adaptive MAC
• 802.11 MAC adapts
rate to minimize
errors
• DSDV routes using
shortest hop-count
paths
– Uses lowest rate to
determine links
• “short” paths can
have less bandwidth
than longer paths!
4Mbps
1Mbps
Cross Layer Design Example:
Rate Adaptive MAC
Plain
Adaptive
Cross Layer Design Example:
Topology Control
• Goal: choose node degree to maximize end-to-end
throughput
• Set transmit power to achieve target-degree
– Short time-scale
• Modify target-degree to increase end-to-end
throughput
– i.e., try to follow gradient to a maxima
– Long time-scale
• Problem: can cause oscillations
– Topology can oscillate between connected and disconnected
states
Cross Layer Design Example:
Topology Control
Topology Control Protocols
• Kawadia and Kumar
–
–
–
–
COMPOW
CLUSTERPOW
Tunneled CLUSTERPOW
MINPOW
• Wattenhofer, et al.
– Angle-based
• Li and Hou
– Directed Relative Neighbor Graph (DRNG)
– Directed Local Minimum Spanning Tree (DLMST)
COMPOW
• Everyone transmits at same power
– Find minimum power s.t. topology remains
connected
• Pros:
– Ensures all links bidirectional
– Allows higher layers to work properly
• Cons:
– Single outlying node causes high-power
CLUSTERPOW
• Run a separate routing protocol at each power level
pi
• Route packets using routing table at minimum pi
where destination is present
• Pros:
– “Clustering” is distributed
– Any base routing protocol works
– Routing is loop-free (power levels monotonically decrease)
• Cons:
– Routing overhead (one per power-level)
– Can’t use initial lower-power hops
CLUSTERPOW
Tunneled CLUSTERPOW
• Recursively lookup path to next hop
– e.g., if D is reachable through N1, search for minpower route to N1, etc.
Tunneled CLUSTERPOW
• Simple recursion is not loop-free
• Solution: Tunnel packet to intermediate hop
MINPOW
• Goal: Route using min-energy route
• Energy cost of using a link at power level p:
– PTotal(p) = PTx + PTxRad(p) + PRx
• Topology:
– Graph is union of topology at all power levels
– Link-cost = minreachable-p(PTotal(p))
– Run DSDV (Bellman-Ford) on resulting graph
• Pros:
– Globally optimal in terms of energy consumption
– Loop-free (just DSDV)
• Cons:
– Not optimal for capacity (but close if PTxRad(p) dominates)
– Does not take into account interference! (i.e., retransmits)
COMPOW/CLUSTERPOW
Throughput vs. Delay
(clustered topology -- mostly 1 hop paths)
COMPOW/CLUSTERPOW
Routing Overhead
Cone-based
• Goal: topology with power efficient
routes
• Assumptions:
– Transmit power dominates energy cost
– Nodes can determine angle of reception
– Trasmit power p(d) = Ω(dx) for x >= 2
• Basic Idea:
– Use min power needed to reach at least
1 node in each cone of 2π/3 around
node (π/2 for optimal efficiency)
– Refine by removing unneeded neighbors
X
Cone-based Properties
• Topology is connected
– Pf. Intuition: consider disconnected u,v with min d(u,v). For
any neighbor w,  > π/3
• Routes are minimum power
– Pf. Intuition: multiple short hops cheaper than one long hop
Cone-based Topology
Max Power
After Phase 1
Final
Cone-based Results
DLMST Motivation
• Goal: Topology formation for nodes with
heterogeneous max power levels
• Problem with Cone-based topology (any
MRNG based method):
DLMST Protocol
• Each node broadcasts HELLO at its max power
• With knowledge of directed graph in its
neighborhood, construct minimum spanning tree
• Pros:
– Connectivity guaranteed
– Node degree bounded by constant (limits interference)
• Cons:
– Links not necessarily bidirectional (can fix, but may sacrifice
global connectivity)
DLMST Results
Average Radius
Average Degree
Topology Control Discussion
• What else besides transmit power
affects topology?
• Is power control a problem in
infrastructure AP networks?
• How can power control affect fairness?
Public Policy:
Spectrum Management
• Spectrum Management Policy Options, Jon
Peha, IEEE Communications Surveys, Fourth
Quarter 1998, Vol. 1, No. 1.
• Approaches to Spectrum Sharing, Jon M.
Peha, Feb. 2005.
• Dynamic Spectrum Policies: Promises and
Challenges, Paul J Kolodzy, Jan 2004.
The Bigger Picture
Staggering Market Statistics
Technology
• 9 million hotspot users in 2003
(30 million in 2004)
• Approx 4.5 million WiFi access
points sold in 3Q04
• Sales will triple by 2009
• Many more non-802.11 devices
Economy
Society
Government
US Spectrum Allocation
802.11
Bluetooth
The Status Quo
• Government licenses spectrum
– By frequency: e.g., for a television channel
– By location: e.g., for the Pittsburgh area
– Only licensees allowed to transmit
• Licenses are temporary
– Allows change in spectrum policy
– New spectrum usually auctioned
– But 99.9% always renewed
• A small number of unlicensed bands
– Industry, Science, and Medicine (prev. slide)
– PCS, NII
– Anyone can transmit (with limitations)
Governing Spectrum Blocks
• Open access: “Flexible use doctrine”
– Let market forces decide applications
– => most value, innovation, competition
• Exclusive access:
– Government chooses application/transmission standard
– => international interoperability, positive “externalities” (e.g., for
police, fire fighters), standardization
Distributing Licenses
• Lotteries
– Avoids political favoritism
– Does not necessarily maximize value
• Auctions
– Tries to maximize value of application
– Can be synchronized to allow buyers to get larger chunks
Alternatives to Licensing
• “Property Rights”
– Treat spectrum same as land
– Allows resale, renting, etc. => opens up secondary
markets for spectrum
– But interference (“trespassing”) on region
boundaries unavoidable
• “Commons”
–
–
–
–
WiFi model: cooperative sharing
Maximize spectrum use if transmission is bursty
Requires some common protocol for cooperation
Requires some altruism
Dynamic Spectrum Management
• Goal: Allocate spectrum more dynamically
– For example, without humans in the loop
• Why? Lots of spectrum is wasted!
– Time of day (some radio stations turn off at night)
– Location (rural areas don’t use all TV frequencies)
– Workload (data applications are bursty)
• Enabling Technology
– Software Defined Radios
– Adaptive Cognitive Radios
– Example: Cordless phones vs. Baby monitors -- manual to
automatic freq. adjustment
Enabling Technologies
• Flexibility
– Can change waveform on the fly (i.e., modulation protocol)
• Agility
– Can change the freq. on the fly (i.e., channel)
• Sensing
– Aware of environmental conditions (i.e., interference)
• Networking
– Can interact with other radios (i.e., ad hoc nets)
Dynamic Policy Options
• Can policy be varied by:
–
–
–
–
–
Transmission duration? (e.g., “TDM”)
RF condition? (e.g., interference sensing)
Short time scales?
Via negotiation between radios?
Impact on environment? (e.g., interference)
• Implementation Options:
– High power beacon to all devices?
– “P2P” networked radio enforcement?
Implementation Challenges
• Quantifying interference
– FCC definition:
“unwanted energy”
• Measurement infrastructure
– Analog to “pollution monitors”
– Dedicated or networked P2P
based?
• Liability policies
– How to punish policy noncompliance
– Do devices need to be certified?
What about software?
• Identity management
– How to identify violators
“Example”
Public Policy Discussion
• How could more dynamic spectrum
allocation impact:
– WiFi Testbeds?
– Community Mesh Networks?
– Mixed Networks?
– Other topics?