Title: Impact of Interference on Multihop Wireless Network Performance Kamal Jain, Jitu Padhye, Venkat Padm
1Impact of Interference on Multi-hop Wireless
Network Performance --Kamal Jain, Jitu Padhye,
Venkat Padmanabhan and Lili Qiu
IEG 5200 Optimization and control in
Communication Networks
- LIU Jian (IE Dept.) and WU Weijie (CSE Dept.)
- --With acknowledgements to authors slides and
Leons slides
2Outline
- Problem raising
- Modeling and bound computing
- Overview of framework
- Protocol model
- Single-path model
- Simulation, results and analysis
- Summary
3Problem raising
- Fundamental issue
- What is the maximum throughput that can be
supported by a given multi-hop wireless network? - Distinction from previous work
- Establishing a general modeling without specific
assumptions
4Outline
- Problem Raising
- Modeling and bound computing
- Overview of framework
- Protocol model
- Single-path model
- Simulation, Results and Analysis
- Summary
5Overview of framework
- Goal To find the maximum throughput
- Architecture of framework
- Model a wired network
- Model a wireless network by adding Interference
constrains - Adding different constrains by giving different
models - Find maximum throughput by solving the
optimization problems
6Wired network modeling
- Formulation for single source-destination model
- LP formulation, easy to solve
7Wireless Network -- Protocol model
- A transmission is successful if
- Receiver stays within the communication range of
the sender - Other nodes that can interfere the receiver are
not sending
- How to express these constrains into the LP
formulations? - Introducing the conception of Conflict Graph
8Conflict graph
- A vertex in a conflict graph F correspond to
the a link in the connectivity graph G. - An edge exist if the two links may not be active
simultaneously.
Conflict Graph (F)
1
2
3
4
9Definitions about conflict graph
- Independent set
- A set of vertices of which no edge exist between
any two of them - Independent vector
- A V-size vector mapping from independent set
- Consider it as a point in a V-dimensional space
1
2
3
One independent set 2,3,4,6 Corresponding
independent vector (0,1,1,1,0,1)
6
4
5
10Definitions about conflict graph
- Independent set polytope
- Convex combinations of independent vectors
- Given independent vectors a1,a2,,ak
- Independent set polytope is Si?iaiS?i1, ?igt0
- Usage vector
- A V-size vector
- each value denotes the fraction of time for each
link
11Constrains under protocol model
- Theorem
- A usage vector is schedulable iff it lies within
the independent set polytope of the conflict
graph. - Corresponding constrains
12Methodology to find optimality
- Difficulty on the constrains
- Given network, sources and destinations
- NP-hard to find all independent sets of a graph
- NP-hard to find or approximate the optimal
throughput - Methodology
- Try to find the lower and upper bound
- When they converge, the value will be the exact
one - When will they converge? Discuss later
13Lower bound
- Equivalent to find a throughput with feasible
schedule - Goal To find the maximum polytope
- How to solve?
- Exponentially expensive to find all maximal
independent sets - To find as many as possible
- Just try to pick easy points Discuss later
- The more we pick, the tighter the bound will be,
and eventually converge to optimal
14Upper bound
- How to lower the upper bound?
- To find as many as constrains
- Particularly total usage of links in a clique lt
1 - Special cases about odd holes and anti-holes
- Anyway, this upper bound may still not be tight
odd holes and anti-holes
15Wireless Network -- Single-path model
- Difference from Multi-path model
- At any node in the network, at most one out-going
edge that has a non-zero flow - Corresponding constrains to add
- Even more difficult to solve integer programs
16Outline
- Problem raising
- Modeling and bound computing
- Overview of framework
- Protocol model
- Single-path model
- Simulation, results and analysis
- Summary
17Simulation, Results Analysis
- Consider a simple 33 Grid as following
9 nodes 24 links
18Simulation, Results Analysis
- We can then present its Conflict Graph like this
1 Conflict 0None
19To get lower bound means
- A simple algorithm to find IS (Independent Set)
- 1. Start with an empty IS
- 2. Add a new vertex to IS iff it doesnt have
an edge to the existed vertices in IS so far - 3. Repeat 2 until we consider all the vertices
- 4. Check to see if we have found this IS
before if not, we add a new constraint to our
LP Otherwise, ignore it. - This is called one unit of effort.
- For the upper bound, we make similar effort to
search for cliques instead of IS.
20Simulation, Results Analysis
- Calculated result of the upper lower bound
Still has a gap
0.5
Converge to optimal value
21Simulation, Results Analysis
Figure1. A Real Neighborhood Map
Figure2. The Connectivity Graph
- Using the basic Single Path Routing, we obtain
the optimal cumulative throughput is 0.5
22Generalizations
- Multiple Source-Destination Pairs
- Assign a connection identifier to each
source-destination pair - Multiple wireless channels
- Easily modeled by introducing M links between
nodes i and j - Multiple radios per node
- Similar to above but may be active simultaneously
23Simulation, Results Analysis
- Possible throughput improvement
- I) Multi-path routing
- II) Double radio range
- III) Two non-overlapping channels
- IV) Two radios per node
- Results
(tend to choose similar path)
(Double the interference too !)
(Double the link capacity)
24Simulation, Results Analysis
More nodes, more interference, throughput
decrease?
- Tradeoff Connectivity V.S. Throughput
Figure1. Connectivity Ratio for 77 grid
Figure2. Normalized per-flow throughput
25Simulation, Results Analysis
- Optimal routing in absence of optimal scheduling
- Four scenarios
3
1
4
2
26Simulation, Results Analysis
- Optimal routing under optimal scheduling
1
2
Value gt 1 1) always better
Table1. Throughput ratios between scenario 1) and
2)
27Simulation, Results Analysis
- Optimal routing in absence of optimal scheduling
3
4
Value vs. 1 3) sometimes worse!
Table2. Throughput ratios between scenario 3) and
4)
28Outline
- Problem raising
- Modeling and bound computing
- Overview of framework
- Protocol model
- Single-path model
- Simulation, results and analysis
- Summary
29Summary
- The paper presented a model and methodology for
computing bounds on the optimal throughput that
can be supported by a multi-hop wireless network. - Key contribution the generality of the
methodology and the conflict graph framework. - The optimal route often outperforms shortest path
route. - The richer connectivity contributed by new nodes
more than offset the increase in traffic load
they cause.
30 31Wireless Network Physical model
- Constrain under the model
- A link exist iff
- Why introduce
- More similar to real network
- Interference gradually increases
- Prospective tighter upper bound
- Weighted conflict graph
32Lower and upper bound
- Lower bound
- Similar to protocol model
- Schedulable set
- Upper bound
- Similar conception of clique
- Additional constrain