Physical Layer Aware Networking

1
Physical Layer Aware Networking

• J.J. Garcia-Luna-Aceves
• CCRG
• Computer Engineering Department
• University of California, Santa Cruz

2
Approach
IMPACT
RESULTS
training of grads and undergrads
analytical models
New models and code used by other groups
capacity analysis
simulations
better models and protocols from other groups
analytical modeling
enable applications of large-scale nets
large-scale simulations
new protocol stack
Prototypes (ucsc testbed)
reports, papers, MURI web page
RESEARCH
3
Wireless Internet in a Slug Box
GloMoSim simulated scenario running in a host
Real-life pictures are captured by WebCam in
real-life and sent through the GloMoSim virtual
network to laptop or PDA as well as
virtual nodes in GloMoSim.
Virtual node in GloMoSim
command center
Image from sensor
4
Need for Cross-Layer Optimization
scheduling establishes links and decides which
nodes are awake needs multicast group
affiliations and routes to destinations of flows
routing needs links for collision-free
transmission of control packets packet
forwarding needs links for collision-free
transmission of data packets Multicasting needs
a convenient topology
topology control determines nodes links that
can be used for certain functions needs links
for collision-free transmission of control
packets, and dissemination of neighborhood data
Signaling to support functions should not be
redundant
5
Key Research Areas

• Help the understanding the fundamental
  limitations to the scaling of ad hoc networks
  with cross-layer optimization (number of nodes,
  energy consumption, bandwidth utilization).
• Study the impact of the physical layer on
  communication protocol stack.
• Modular protocol stacks to bridge the gap between
  the applications of large ad hoc networks and the
  new hardware available with ST coding and other
  technologies.
• Complement ing MURI research with other ongoing
  research work at UCSC

6
Modular Protocol Stack
7
Details
8
Prior Results on Network Capacity

• Definition A source-destination throughput of
  ?(n) bits/sec is feasible if every source node
  can send information at a rate of ?(n) bits/sec
  to its destination.

• Gupta and Kumar (for static networks)

• Grossglauser and Tse (Multiuser diversity
  One-copy two phase packet relay to nearest
  neighbor strategy for mobile networks)

9
Network ModelNode Trajectories Are iIID
10
Contributions to Date

• We present multiuser diversity with multi-copy
  two phase packet relay to close neighbors
  strategy for mobile networks where

• For fixed n

• Interference analysis

11
MURI Research

• How can we improve the throughput performance by
  sending packet via least resistance paths?
• We will need cross layer optimization!

12
Why Do We Need Analytical Models?

• Analytical Models
• Aim to cover different scenarios general
  behavior!
• Quick answers for the impact of different
  parameter values on system performance
• Upper/lower bounds
• Insights help in the design
• Physical layer issues at least as accurate as in
  simulations

• Simulations
• Specific to each scenario and setup
• Results for each parameter value of interest
• Statistical fitting not a trivial task
• Many physical layer features not readily
  available
• Poor physical layer implementations
• How far can we go?

13
Modeling ad hoc networks What are the challenges
?
Point-to-point network
Ad hoc network
X

• Channel shared by single pair of nodes
• Reliable, stationary medium
• Well-defined topology

• Channel shared by many nodes channel access
  protocols !!!
• Unreliable, non-stationary medium
• Topology not a boolean function

14
Hidden-Terminal Problem

• Well-known but poorly understood!

B
C
A
packets collide at B !!!
A does not sense Cs transmissions to B.
15
Multihop Networks
Interference is network-wide!
16
Modeling the Effect of the PHY Highlights

• Framework for any MAC protocol in ad hoc networks
• Focus on PHY / MAC layer interactions
• No assumptions on spatial probability
  distributions or specific arrangement of nodes
• Individual (per-node) performance metrics for any
  given network topology (node location) and radio
  channel model
• Analytical results faster than in simulations (a
  few seconds compared to hours of simulation).
• Linear model that provides remarkable correlation
  with simulation results.

17
Previous Work

• Single-hop (mostly) or weak-interactions
  approach (to avoid interference from distant
  nodes)
• Scheduling rates are independent Poisson point
  processes
• Packet lengths exponentially distributed and
  independently generated at each transmission
  attempt backoff retransmissions ignored!
• Instantaneous acknowledgments
• Error-free Links
• Assumptions on spatial distributions (e.g.,
  Poisson)

18
Modeling Rationale

• Focus on the essentials of MAC and PHY layers
• PHY ensure that frames are correctly received
• MAC scheduling discipline to share the channel
• MAC- and PHY-layer dynamics tightly connected
• MAC/PHY interactions depend on connectivity among
  the nodes
• Network topology is key!
• Model each layers functionality in probabilistic
  terms
• PHY successful frame reception probability
• MAC transmission probability
• (i.e., a scheduling rate)

19
Application Modeling IEEE 802.11

• Based on the works by
• M. Carvalho and J. J. Garcia-Luna-Aceves, Delay
  Analysis of IEEE 802.11 in Single-Hop Networks,
  Proc. ICNP, Atlanta, 2003.
• G. Bianchi, Performance Analysis of the IEEE
  802.11 Distributed Coordination Function, IEEE
  JSAC, 2000.

20
Application Modeling IEEE 802.11

• Per-node performance metric throughput

Simulator used Qualnet 3.5
21
Percentage Prediction Error
Sample topologies
Histogram over 10 random topologies (100 nodes)
22
Modular Protocol Stack
23
Channel Access Protocolsgt Comparing with CSMA
and CSMA/CA
24
CA vs Dynamic Scheduling(Analytical Results)

• Dynamic scheduling is much better than CA
  however, it is not enough!

25
MURI WorkNeed for Flow Activation

• With link or node activation, the opportunities
  for collision-free transmissions are not related
  to the flows traversing the MANET.
• Channel may go without use if there is no traffic
  to be transmitted by node A or over link (A,B)
  when the entity is activated.
• We need data packets to obtain access to the
  channel quickly and without conflicts!

26
Flow-Aware Scheduled Transmission (FAST) Protocols

• Run the anticipatory collision resolution using
  flow identifiers as the entities competing for
  channel access.
• A flow ID can be
• sourcedestinationseq.number assigned by source
• Use a neighbor protocol to communicate competing
  flows.
• Flows can be unicast or broadcast, single hop or
  multihop.
• Some other details Distance of a node to flow
  source, hidden terminals.
• Note This assumes routing information is
  available to guide the dissemination of flow
  information (along shortest paths to flow
  destinations).

27
MURI WorkRouting Issues

• Routing protocols are monolithic
• One flavor of signaling for all classes of
  destinations
• One flavor of routes (single path) for all
  classes of traffic to destinations.
• Routing layer in MANETs assumes that routing
  takes place over a given topology, just like
  Internet routing protocols like OSPF and RIP do.
• The existence of radio connectivity does not
  imply the availability of a logical link in a
  MANET.
• We need FAR MORE!

28
Flow Adaptive Routing (FAR)

• Goal is Scaling and QoS Support
• Develop routing techniques that are node-centric
  (no clusters) and adapt dynamically to the flows
  in the network.
• How a routing table entry for a destination is
  obtained and maintained is a function of the type
  of flow towards the destination.
• Proactive and on-demand mechanisms used according
  to flow types.
• Different flows are given resources (paths)
  according to their types and priorities.
• Routing works in coordination with scheduling.

29
Integrated Routing and Multicasting

• Each node transmits group announcements when
  distances to cores change.
• Group announcement states one or multiple
  multicast groups, the core of the group and its
  distance to the core.
• Each node in a group transmits join announcements
  periodically to indicate it is active in a group.
• A well-known sink is its own core.
• A source that is not in group sends unicast
  packets towards core of group, and packet is
  multicast from the first node hit that is in the
  group.
• Nodes communicate their network resistance for
  each destination to their neighbors.

30
Integrated Routing and Multicasting
Each common node keeps paths to the cores of
groups and well-known nodes. Paths to common
nodes are found on demand. Much of the traffic in
sensor nets is to groups and common nodes!
31
(No Transcript)