SLIDE 1
Using Collision-Free Scheduling: Dream or Reality?
Gil Zussman Department of Electrical Engineering Columbia University
Theory and Practice in Wireless Networks, USC May 2008
SLIDE 2 Medium Access Control
Nodes need to coordinate the access to the medium
- Transmission time, power, channel, rate, etc.
Distributed
- Random Access (Aloha, CSMA, etc.)
- Collision Free (TDMA, FDMA, etc.)
In the wireline domain (Ethernet)
- Random Access (CSMA/CD) Collision Free (switching)
In the wireless domain
- Is collision free doable?
- If yes, what can be achieved? (throughput, delay, fairness, energy
efficiency)
2 3 5 6 4 7 8 1
SLIDE 3
My First Random Access System (Winter 1990/1)
Army/Navy “Portable” Radio Communication System (AN/PRC-77) Suffered from collisions, interference, and back pain (20 Lb.)
SLIDE 4 10 iRobot Roombas with
IEEE 802.11g
Suffers from collisions
and interference
your apartment * Jointly with J. Reich, V. Misra, and D. Rubinstein
Our Current Random Access System* (2008)
SLIDE 5
Multihop Wireless Network
Usually no predefined topology no “Links” and no “Neighbors” Not really “Disks”
2 3 5 6 4 7 8 1
SLIDE 6
Wireless Networking Technologies
ZigBee
SHORT < RANGE > LONG LOW < DATA RATE > HIGH
Personal Area Network Local Area Network
Bluetooth 802.11b 802.11a/g/n
Metropolitan Area Network / Cellular
802.16 WiMedia
The systems we have been experimenting with are mostly Random
Access systems
LTE UWB Military / Public Safety
SLIDE 7 IEEE 802.16 (WiMax) Mesh
Provides the backbone between the Base Stations
- Can be used in Rural Areas
- Nodes coordinate with their two hop neighborhood
- Coordinate a collision free schedule
SLIDE 8 Military Systems
Joint Tactical Radio System (JTRS) and similar systems are
being developed
- Based on Software Defined Radio
There is no “one size fits all” solution
- MANET between fighter planes ≠ MANET between infantry soldiers
- Many waveforms
- Some will use Random Access and some Collision Free
SLIDE 9
Collision Free vs. Random Access
Have been around for decades Different systems need different approaches (I believe) we will keep seeing both Another important dimension - Theory vs. Practice
SLIDE 10
Theory vs. Practice
Fundamental understanding of wireless networks Practical solutions to real-world networking problems Interference Graph Models Realistic Channel Models (SINR-Based) Collision Free Random Access
SLIDE 11
Ongoing Research - Collision Free/ Interference Graph
Network Model (Tassiulas and Ephremides, 1992) Time-slotted system Stochastic arrivals – i.i.d. process with arrival rates λij (are not known
in advance)
Only a subset of the links can be activated simultaneously, due to
interference
λ12 , λ14 , λ16
2 1 3 5 6 4 7 8
λ21 , λ24 , λ28 λ6i , ...
SLIDE 12 Ongoing Research - Collision Free/ Interference Graph
Developing distributed algorithms
- Based on the centralized framework of
Tassiulas and Ephrimedes (1992)
Tradeoffs between decentralization, complexity, throughput, delay,
fairness, and the effects of topology, and interference
Decentralization/ Complexity Throughput/ Delay Interference Topology ?% 100% O(?) O(??)
SLIDE 13 Randomization approach [Modiano, Shah, Zussman, 06]
Randomized scheduling framework that achieves 100%
throughput
- Based on a result of Tassiulas, 1998
Comparable complexities to the deterministic distributed
greedy algorithms that achieve fractional throughput
SLIDE 14 Partitioning Approach [Brzezinski, Zussman, Modiano, 06,08]
Identified graphs in which distributed algorithms achieve
100% throughput
- Based on the notion of Local Pooling (Dimakis and Walrand, 2006)
Examples
- Trees under any interference
degree
secondary interference
SLIDE 15 What’s Next?
The one time scheduling problem is already hard Approaches for the general problem
- Randomized Scheme
- Local Pooling
Interference Graph Models Realistic Channel Models Collision Free Random Access
?
(SINR > γ)
