Multi-level Hashing for Peer-to-Peer System in Wireless Ad Hoc Environment

1
Multi-level Hashing for Peer-to-Peer System in
Wireless Ad Hoc Environment

• Dewan Tanvir Ahmed and Shervin Shirmohammadi
• Distributed Collaborative Virtual Environments
  Research Laboratory
• School of Information Technology and Engineering
• University of Ottawa, Ottawa, Ontario, Canada
• dahmed, shervin_at_discover.uottawa.ca
• Pervasive Computing and Communications
  Workshops(PerComW'07)
• ?????

2
Outline

1. Introduction
2. Multi-level hashing in wireless ad
   hocpeer-to-peer systems
3. Wireless ad hoc p2p system node join
4. Wireless ad hoc p2p system content publishing
5. Wireless ad hoc p2p system content discovery
6. Simulation
7. Conclusion

3
Introduction (1/2)

• Mobile peer-to-peer (MP2P) system faces more
  constraints as compared to wired peer-to-peer
  system. Some problems in this network are
• Scarcity of bandwidth
• Short lifetime of the nodes due to power
  constraint
• Dynamic topology caused by the mobility of nodes.
• These constraints give hard challenges to the
  researchers in the various aspects of the
  peer-to-peer system in ad hoc environments such
  as
• Routing
• Content discovering
• Data retrieval
• Caching

4
Introduction (2/2)

• Smart content discovery is the heart of peer-to-
  peer system. In ad hoc network, routing is
  expensive so it requires more effective content
  discovery process.
• We apply chord directly on top of wireless ad hoc
  peer-to-peer system.
• We use two-level distributed hashing scheme.
• First level hash maps a content request to a
  region that greatly reduces routing overhead.
• Second level hash discovers content location
  after redirecting the request to that region.

5
Multi-level hashing in wireless ad
hocpeer-to-peer systems (1/2)

• The entire system consists of a set of home
  regions where peers in close proximity form home
  region.
• Each home region has a representative that plays
  a supporting role to make the system operational.
  

6
Multi-level hashing in wireless ad
hocpeer-to-peer systems (2/2)

• These representatives may change over the time
  considering the ad hoc issues of the system.
• Home region is identified by the Home-Region-ID.
• The representative functions as a rendezvous
  point for incoming requests and also provides
  useful information to its local peers.

7
Wireless ad hoc p2p system node join (1/3)

• We use a pair, ltHome-Region-ID, Node-IDgt, to
  identify a node.
• Hash function generates Node-ID
• Node-ID Hl(Node-Address)
• Home-Region-ID depends on its location that may
  change over time.

8
Wireless ad hoc p2p system node join (2/3)

• Two ways to learn about home region
  representative.
• The home region representative periodically
  broadcasts its own identity (address) and
  Home-Region-ID.
• Joining node explicitly floods Home-Region-Search-
  Request.
• Home-Region-Search-Request is a controlled flood
  not flooded to the entire system.
• An intermediate node does not forward the request
  when it already has the answer of the request.

9
Wireless ad hoc p2p system node join (3/3)

• The requester may get multiple responses based on
  its location.
• It marks minimum hops away response.
• The joining node confirms the representative for
  its inclusion to the system.
• It gets the identity of its successor and
  predecessor from this representative.
• The joining node determines its Node-ID using the
  hash function Hl.

10
Wireless ad hoc p2p system contentpublishing
(1/5)

• keyr Hr(Content-Name )
• Map the content name to a home region.
• keysHl(Content-Name)
• Determine the successor in a particular region,
  i.e. the actual node where content information
  has to be stored.

11
Wireless ad hoc p2p system contentpublishing
(2/5)

• keyr, is used to determine the home region where
  content has to be published.
• The content publisher first builds a tuple
  consisting of ltContent-Name, Own-Addressgt and
  then asks local representative to store the
  tuple.

12
Wireless ad hoc p2p system contentpublishing
(3/5)

• Local representative transport the tuple to the
  foreign region. But it may happen that content
  has to be stored in the local region based on the
  keyr.

13
Wireless ad hoc p2p system contentpublishing
(4/5)

• successor(keys)
• the node identifier of the first actual node
  following keys around the circle clockwise.
• At a particular region, the successor of the
  content is determined using keys and the
  representative initiates tuple storing process.

14
Wireless ad hoc p2p system contentpublishing
(5/5)

• The index is distributed over the nodes at
  random. One of the shortcomings in this
  environment is nodes ad hoc nature.
• To recover from such unwanted events protocol
  deploys multiple hash functions to store tuple in
  multiple nodes.

15
Wireless ad hoc p2p system contentdiscovery
(1/8)

• Hashes to get keyr using Hr(Content-Name).
• This key represents the home region where the
  content information is stored.
• If the content is not in local region, the
  request is delegated to the foreign
  representative.
• Node gets the address of foreign representative
  with the help of local representative.
• The protocol allows local representative to
  broadcast home regions addresses in its periodic
  keep-alive messages.

16
Wireless ad hoc p2p system contentdiscovery
(2/8)

• We applied chord like content searching process
  with modified finger table in local region.
• Lookup can proceed as follows.
• Based on keyr, either the initiating node or the
  foreign representative sends a packet to its
  successor containing requesters physical address
  and the key, keys.
• The packet is circulated around the logical
  circle until it locates the successor or it fails.

17
Wireless ad hoc p2p system contentdiscovery
(3/8)

• Chord uses finger table to speed up searching
  process. If the hash address space is too large
  as compared to the number of peers in the system
• The most of the entries in finger table point to
  the same node, and search continues in linear
  way.
• Let hash key is 128-bit long. Most entries of the
  finger table point to the same successor.
• We propose two approaches to construct finger
  table for wireless ad hoc environments.

18
Wireless ad hoc p2p system contentdiscovery
first approach (4/8)

• In the first design, the finger table is slightly
  different from chord.
• The finger table has m entries, indexed by 0
  through m-1. It is divided into two halves
• 0 to m/2
• m/2 to (m-1).

19
Wireless ad hoc p2p system contentdiscovery
first approach (5/8)

• The values of the fields for entry i at node k
  for the first half of the table are
• First half table covers k to k 2m / 2-1 address
  space.

start the physical address of successor(start)
k 2i mod 2m, where 0 i lt m/2 Physical address of successor(starti)
20
Wireless ad hoc p2p system contentdiscovery
first approach (6/8)

• step-size
• a big number to jump more rapidly.
• The entries of the second half of the table are
  incremented by the step size starting from k
  2m / 2-1 .
• Similarly we have corresponding physical
  addresses of the successors.

21
Wireless ad hoc p2p system contentdiscovery
first approach (7/8)
22
Wireless ad hoc p2p system contentdiscovery
second approach (8/8)

• The values of the fields for entry i at table of
  node k are
• a is a constant
• One possible choice for the value of a is the
  index of finger table.
• i.e. it increase at a rate of ii 11, 22, 33.
• n is n-bit hash value.
• These approaches perform well when the number of
  peers is too small as compared to the address
  space otherwise we prefer the usual one.

Start the physical address of successor(start)
(k ai) mod 2n ,where m/2 i lt m Physical address of successor(starti)
23
Simulation (1/8)

• 8000 10000 peers are randomly deployed in the
  environment
• Each peer has 0.2 movement probability.
• Nodes are allowed to move at any direction.
• 16 fixed adjacent regions.
• Not all nodes are involved in peer-to-peer
  system.
• If we have 8000 peers, system has more than 8000
  nodes.
• Peers are considered alive during the simulations
  but helping nodes (intermediate ad hoc nodes that
  provide routing service) are not all time.
• The request may not end with a success.

24
Simulation (2/8)
25
Simulation (3/8)

• Applied chord directly on top of wireless ad hoc
  network (Setup 1) .
• Estimated search cost and routing cost of chord
  for peer-to-peer system in wireless ad hoc
  network.
• Search cost
• How many peers are involved in a particular
  content discovering process
• Routing cost
• All ad hoc nodes that are involved in content
  discovering process including the forwarding
  nodes.

26
Simulation (4/8)

• Chord in wireless ad hoc network is not suitable
  as a peer-to-peer system because of its high
  routing cost.
• If the number of peers is too small as compared
  to the hash address space, finger table has
  little contribution in content discovering
  process.

27
Simulation (5/8)

• Two-level hashing presented in this paper is
  tested and compared with chord for wireless ad
  hoc p2p system.
• Simulation environment(Setup 2)
• For the sake of fair experiment, we ignored
  finger table in both approaches. This is because
  we can deploy any variant of finger table in both
  approaches.

28
Simulation (6/8)

• On an average, chord requires 10 to 12 times more
  searching cost than our two-level hashing
  approach.
• Content searching cost is reduced as we divide
  the entire topology into multiple small regions
  and first level hash makes it possible to explore
  the content into that small region.

29
Simulation (7/8)

• Chord requires 50 times more routing cost than
  two level hashing scheme on an average.
• Chord is a logical topology. It has no knowledge
  of the physical structure.
• Request is redirected to a node that is logically
  close but physically far away.

30
Simulation (8/8)

• It is also true in two-level hashing approach
  when it enters in a local region.
• But its logical ring is composed of close
  proximity nodes as compared to one level hashing
  in chord.
• Reduces routing cost to a large extent.
• Proves the requirement of multi-level hashing.

31
Conclusion(1/1)

• We present a two-level hashing scheme for
  wireless ad hoc peer-to-peer system.
• It shows that the performance of the system will
  be increased once we have multi-level hashing in
  terms of searching cost and routing overhead.
• Modified finger tables works well when the number
  of peers is too small as compared to the hash
  address space.