Download Medium Time Metric

High Throughput Route
Selection in Multi-Rate Ad Hoc
Wireless Networks
Baruch Awerbuch, David Holmer, Herbert Rubens
Szikszay Fábri Anna, ELTE IK Prog.terv.mat.
2009.05.05
Table of Contents








Basic terms
Network model
Traditional route selection techniques
General model of attainable throughput
MTM: Medium Time Metric
Advantages
Discussion
Summary

OAR


receiver based approach
allows high-rate multi-packet bursts


to take advantage of the coherence times of good channel
condition
bursts also dramatically reduce the overhead at high
rates
Basic terms

Ad hoc wireless network:



Multi-rate network:


Decentralized
each node is willing to forward data for other
nodes, and so the determination of which nodes
forward data is made dynamically based on the
network connectivity
which allocates transmission capacity flexibly to
connections
OAR: Opportunistic Auto Rate protocol
Network model

Network assumptions





the OSI/ISO layer is capable of operating using
multiple rates
the ISO/OSI MAC layer is capable of selecting
the rate used by the physical layer
the MAC layer is capable of providing
information to the ISO/OSI network layer that
indicates the selected rate
The network layer can then use this information
to improve its routing decisions
Demonstration: information from lower
layers can be utilized to enhance overall
performance
Traditional route selection
techniques
1. Minimum hop path
 hop count: route selection criteria



minimizes the total number of transmissions
required to send a packet on the selected
path
in single-rate wireless networks: OK


Most ad hoc routing protocols
every transmission consumes the same amount
of resources
multi-rate networks:

tendency: pick paths with both low reliability and
low effective throughput

Throughput loss

Multi-rate wireless networks




the selection of minimum hop paths → paths where
the links operate at low rates
Shortest path contains fewer nr of nodes
To cover the same distance → longer links →lower
channel quality (lower rates) → low throughput
Shared medium: degrades the path of other flows in
the network

Transmission of a packet at low link speed: takes TIME..

Reliability loss



Multi-rate wireless devices are designed to deal
with connectivity changes (mobility &
interference)
In 802.11b prot.: 2 nodes move in opposite
directions→ link speed drops
Tendency: lowest link speed path

No chance for auto rate protocol to deal with channel
quality fluctuations
2. Shortest Widest Path
 selects the shortest path from the set of
paths that have the fastest bottleneck link
 commonly used routing criteria in wired
networks



the total throughput of a path is directly related
to the speed of the bottleneck link (each link in
the path operates independently)
often used when high throughput is required
Inappropriate for wireless networks


In wireless networks, individual links do not
operate independently of one another
Individual transmissions affect a large area




compete for medium time
other transmissions along the same path
transmission in the same geographical area
does not consider the speed of links other than
the bottleneck

even though these links my affect the bottleneck link!

Conclusion:the two paths equal
(throughput)

each have equal bottleneck links → selection:
which path contains the fewest hops
General Model of Attainable
Throughput



Difficulty: in modeling the complex
environment of wireless multi-hop networks
We ignore packet scheduling issues and
consider a steady-state flow model
The model




each network edge may be fractionally shared
by several flows
the sum of shares cannot exceed 100%
the transmission graph: G(V, E, ρ)
transmission rate to each transmission edge ρ :
E → R+


G can be directed
the transmission rate in the reverse direction of a bidirectional edge may be different than that in the
forward direction


different node configurations and asymmetric channel effects
The interference graph: G(V˜ , E˜)




the vertices of the interference graph to be the edges
of the transmission graph: V˜ = E
((a, b), (c, d)) ∈V˜ if (a, b), (c, d) ∈ E and if a
transmission on (a, b) interferes with a transmission
on (c, d)
modeling the interference graph: difficult
interference neighborhood of any given edge (u, v) as
follows. χ(u, v) = (u, v) ∪ ((x, y) : ((x, y), (u, v)) ∈ E)


a set of i flows: each φi originates from source si and
is sinked by receiver ri .
we can represent each flow as a sum of path flows.
Each path flow φij exists only on πij (path)
 each edge (u, v) in the transmission graph
 the sum of the fractional shares used by all flows in
the interference neighborhood of (u, v) must be less
than or equal to 100%.
 this is a more complicated version of the classic edge
capacity flow constraint.


Linear Programming (LP) methods are required
to achieve an optimal throughput solution
Theorem 1:

In the case of a complete interference graph in
the stated multi-rate ad hoc wireless network
model, a routing protocol that chooses a path
that minimizes the sum of the transmission
times minimizes network resource consumption,
and maximizes total flow capacity.
Medium Time Metric


It is designed to allow any shortest path
routing protocol to find throughput optimal
routes assuming full interference
assigns a weight to each link


proportional to the amount of medium time used
by sending a packet
Weight of a path

Sum: proportional to the total medium time
consumed


packet traverses the whole path
→ shortest path protocols that use the
medium time metric find paths that
minimizes the total transmission time


Assumption: full interference
A general optimal algorithm must





monitor the medium time utilization at every
node in the network,
disseminate that information (to aid routing
decisions)
unnecessary to use multiple paths
simultaneously
Multiple paths available: random choice &
exclusive usage
Using additional paths: no advantage

Computing link weights



MTM: paths that minimize the total consumed
medium time should be selected
Uses existing shortest paths protocols
Assign WEIGHT to LINKS


Medium time consumed ~ package sending
Possible solution

Inverse rate scheme (Cisco)




In wired networks: usage of MTM no advantage
There is no transmission interference in wired nws
 Wires are isolated :)
BUT: prediction about medium time consumed is sometimes
wrong (because of MAC overhead)
MAC overhead

The solution:



To induct a package size dependency into the
protocol
Transmission of a small packet is dominated by the
MAC overhead & is almost the same (regardless to
link rate)
MTM would use different light weights for different
package size


easy to implement in link state protocols
 topology information (to compute alternate routes using
different sets of weights)
More difficult for distance vector protocols

additional communication overhead for each additional set
of weights

An implementation of the MTM for a distance
vector protocol

It is tuned for the dominant packet size





using link weights ~ to the medium time
 used by packets of the tuned size
Larger packets: longer path with even higher rate links
Smaller packets: paths that are shorter but with lower rate
links
the tuned packet size was chosen (1500 byte IP packet)
OAR protocol significantly changes the MAC layer
packet exchange


the expected medium time consumed by a packet at a given
rate changes significantly
MTM weights must be calculated to match the change in
consumed medium time

Simplicity

shortest path metric




Advantages
it can be incorporated into existing distance vector or
link-state protocols
the majority of existing wireless ad hoc routing
protocols fall into these categories
MTM protocols only need to track changes
in link rates
MTM paths naturally avoid low-rate links



nodes connected by a high-rate link consider
distance before the link breaks
nodes move apart, the auto rate protocol
reduces the link speed
proactive routing protocols: update their paths
→ avoid path failures

by continuously switching to higher rate links.
Discussion

link rates by definition change faster than
link connectivity


some routing protocols may consume more
overhead when using MTM when compared
with min hop
distance is the dominant factor that
determines the link rate

even in the worst case, the MTM metric should
only change a constant amount more than
connectivity

better than traffic sensitive routing,

traffic loads change much faster than either link rates or link
connectivity

The MTM selects paths that have a greater
number of hops than the minimum




higher rate hops: less total medium time than
the minimum number of hops
BUT: increased number of senders could cause
other detrimental effects: packet drop
When the density of the network is low: topology
sparsely connected
few choices for routing protocols to select from

MTM and min hop will tend to pick the same path

Node density ~ increased throughput
Summary





general theoretical model of the attainable
throughput in multi-rate ad hoc wireless
networks
MTM is derived from a detailed analysis of
the physical and medium access control
layers
Selects optimal throughput paths and tends
to avoid long unreliable links
Minimizes the total medium time consumed
sending packets from a source to a
destination
This results in an increase in total network
throughput (20% to 60%)