EnergyEfficient Caching Strategies in Ad Hoc Wireless Networks

1
Energy-Efficient Caching Strategies in Ad Hoc
Wireless Networks
EECS 600 Advanced Network Research, Spring 2005
Instructor Shudong Jin March 23, 2005
2
Caching in Many Networks

• Parallel and Distributed systems
• Reducing communication overhead, sharing,
  consistency, etc
• Internet (Web) Caching
• Reducing latency/traffic, dynamic contents,
  proxies, etc
• Content Delivery Networks (CDN)
• Broadly, also Internet caching
• Multimedia caching
• Improve QoS/timeliness, reduce traffic/startup
  latency, etc
• Many other applications of caching techniques

3
Caching in Ad Hoc Networks?

• Necessary?
• Scarcity of communication bandwidth
• A key issue is to satisfy user requests with
  minimum service delay.
• Limited energy resources
• The energy expended for transferring information
  across the network has to be minimized.
• Possible?
• Internet-based server provides content to a set
  of nodes with large storage, i.e., they are
  caches.
• Beneficial/Feasible?
• Whenever access latency and energy cost of data
  transfer are high, the best approach is to cache
  the requested information at a limited number of
  nodes distributed across the network.
• Need wise caching decisions

4
This Paper

• Problem formulation
• An integer linear program, the same as a special
  case of the connected facility location problem,
  which is known to be NP-hard.
• A polynomial time algorithm which provides a
  sub-optimal solution
• The proposed algorithm applies to any arbitrary
  network topology and can be implemented in a
  distributed and asynchronous manner.
• In the case of a tree topology, the algorithm
  gives the optimal solution.
• In the case of an arbitrary topology, bound 6
• Performance comparison of our algorithm against
  three candidate caching schemes

5
System Model and Assumptions

• Multi-hop network G(V,E), with one server in V.
• Dissemination phase put copies in nodes (zk1)
• Access phase node k requests a copy with pk, and
  dk is the distance between k and a copy (any)
• Bidirectional links and constant link cost
• Insignificant storage cost
• Other hidden assumptions (some are very
  unrealistic)
• Consider aggregate cost (not individual nodes)
• Network connected
• Stationary period long enough
• No cost to discover the caches

6
Cost Functions

• Energy cost
• Dissemination cost Kdiss-energy
• Access cost Kacc-energy
• Latency
• Latency Klatency
• Total cost (awkward weighted sum of different
  units)

7
Problem Formulation

• Did they consider the piece of information can be
  cached by intermediate nodes while being
  requested by a node?

8
Facility Location Problem

• Single facility location
• Find a location (say x-y coordinates) that
  minimizes a weighted sum of distances to each of
  several locations.
• Connected facility location problem
• An existing facility is given, along with a set
  of locations at which further facilities can be
  built. Every location k is associated with a
  service demand, denoted by pk, which must be
  served by one facility. The cost of serving k by
  using facility j is equal to pkckj, where ckj is
  the cost of connecting k to j. Besides selecting
  the sites to build the facilities, we also want
  to connect them by a Steiner Tree. Connecting the
  facilities incurs a cost which is proportional to
  the weight of the Steiner Tree. The objective is
  to find a solution which minimizes the total cost.

9
Greedy Solution Tree Model

• Basic idea push a copy down a branch as long as
  it is beneficial to do so
• Let us use an example on the board
• Easy to understand its optimality

10
Greedy Solution General Networks

• Basic idea push a copy further from the server,
  to a node, such that it is most profitable.
• Example shows Ne, Be, Re

11
Skip Section 5,6,7
12
Algorithms Compared Numerically

• No caching (NC)
• Depth caching (DC)
• Push copies to all nodes within h hops
• Flooding (FLD)
• POACH
• Why not show the optimal solution?

13
Numerical Results (1)
14
Numerical Results (2)
15
Weaknesses of this Work

• Less realistic assumptions
• E.g., insignificant storage cost in nodes?
• Lack of novelty
• Similar work in Internet environment, but use a
  different cost function
• Total cost a linear combination of two different
  ones
• Search cost is ignored
• How to locate a copy?
• Mobility versus caching effectiveness not
  exploited
• Intuitions behind the Greedy algorithm not clear

16
Comments from Students