Effective Replica Allocation

1
Effective Replica Allocation in Ad Hoc
Networks for Improving Data Accessibility
Takahiro Hara
(Proc. IEEE Infocom 2001,pp1568-1576)
Presented by Mingsheng Peng
2
Contents

• Why Data Replication
• Related Work
• System Model
• Replica Allocation Methods
• Simulation Model
• Conclusion

3
Why data Replication?

• Ad hoc networks temporarily constructed by only
  mobile hosts.
• Mobile host plays the role of a router, even if
  source and destination are not in range, data
  packets are forwarded by relaying
• Since hosts move freely, disconnections occur
  frequently, this causes frequent network division.

4
Why data Replication?(contd...)

• Example, if some link goes down and the
  network is split
• Nodes on the right cannot access D2
• Nodes on the left cannot access D1

5
Why data Replication?(contd...)

• A possible solution is by replicating data items
  at mobile hosts which are not the owners of the
  original data.

6
Related Work

• Ad hoc network routing protocol Such as
  DSDV,AODV,DSR,ZRP,CBRP
• can only improve the connectivity among MHs which
  are connected to each other,
• but cannot do anything when the network is
  divided as in the case in Figure 1.
• Distributed database systems
• data Replication in database helps in reducing
  response time
• since failures occur infrequently, a small number
  of replicas is sufficient
• Mobile computing
• mobile hosts access databases at sites in a
  fixed network, create replicas on mobile hosts
• address issues of maintaining consistency with
  low communication costs
• assume only one-hop wireless communication

7
System Model

• The system environment is assumed to be an Ad-hoc
  network where
• mobile hosts access data items held by other
  mobile hosts (single or multiple hops)
• each mobile host creates replicas of the data
  and maintains the replicas in its memory
• data item available if it is present locally or
  if it is available at one of the neighbors
• Assumptions
• unique host identifier Mj (set of all mobile
  hosts M M1, M2,, Mm)
• unique data identifier Dj (set of all data
  items D D1, D2,, Dm)
• Assume all data items are of the same size
• each host has a memory space of C data items for
  replicas (excluding the space for holding
  originals)
• data remains the same and does not change
  (simplifying assumption)
• access frequencies of data items from each
  mobile host is known and does not change

8
Replica Allocation Methods

• Approach
• replicas are relocated in a specific period
  (relocation period)
• replica allocation is determined based on the
  access frequency and network topology

9
Three replica allocation methods

• Three replica allocation methods differ in
  emphasis put on access frequency and network
  topology.
• SAF (Static Access Frequency) only the access
  frequency to each data item is taken into
  account.
• DAFN (Dynamic Access Frequency and
  Neighborhood) The access frequency to each data
  item and the neighborhood among mobile hosts are
  taken into account.
• DCG (Dynamic Connectivity based Grouping) The
  access frequency to each data item and the whole
  network topology are taken into account.

10
SAF(static Access Frequency)

• Each host creates replicas in descending order of
  access frequencies
• Advantages
• No control information regarding replicas need
  to be exchanged
• Once each host has its all necessary replicas,
  there is no more replica relocation.
• Low overhead and low traffic
• Disadvantages
• hosts with similar access characteristics have
  the same replicas, but a MH can access data items
  held by other connected MHs,and it is more
  effective to share many kinds of replicas among
  them.Thus it gives low data accessibility when
  many hosts have the same or similar access
  characteristics.

11
SAF example
12
DAFN(Dynamic Access Frequency and Neighborhood)

• The algorithm of this method is as follows 
• 1)  At a relocation period, each mobile host
  broadcasts its host identifier and information on
  access frequencies to data items. After all
  mobile hosts complete the broadcasts, from the
  received host identifiers, every host shall know
  its connected mobile hosts.
• 2)  Each mobile host preliminary determines the
  allocation of replicas based on the SAF method.
• 3) In each set of mobile hosts which are
  connected to each other, the following procedure
  is repeated in the order of the breadth first
  search from the mobile host with the lowest
  suffix (i) of host identifier (Mi). When there is
  duplication of a data item (original/replica)
  between two neighboring mobile hosts, and if one
  of them is the original, the host which holds the
  replica changes it to another replica. If both of
  them are replicas, the host whose access
  frequency value to the data item is lower than
  the other one changes the replica to another
  replica. When changing the replica, among data
  items whose replicas are not allocated at either
  of the two hosts, a new data item replicated is
  selected where the access frequency value to this
  item is the highest among the possible items.

13
DAFN(Dynamic Access Frequency and Neighborhood)

• Eliminates replica duplication among neighboring
  hosts
• The above procedure is executed every relocation
  period
• Overhead and traffic is much higher than SAF
• Does not completely eliminate replica
  duplication
• If network topology changes during the execution
  of this method, replica relocation cannot
  completed

14
DAFN example
15
DCG(Dynamic Connectivity based Grouping)

• Biconnected component A maximum partial subgraph
  which is still connected if one of the vertices
  is removed (high stability!)
• The algorithm is as follows 
• 1) At a relocation period, each mobile
  host broadcasts its host identifier and
  information on access frequencies to data items.
  After all mobile hosts complete the broadcasts,
  from the received host identifiers, every host
  knows the connected mobile hosts.
• 2) In each set of mobile hosts which are
  connected to each other, from the mobile host
  with the lowest suffix (i) of host identifier
  (Mi), an algorithm to find biconnected components
  is executed. Then, each biconnected component is
  put to a group. If a mobile host belongs to more
  than one biconnected component, i.e., the host is
  an articulation point, it belongs to only one
  group in which the corresponding biconnected
  component is first found in executing the
  algorithm.
• 3) In each group, an access frequency of the
  group to each data item is calculated as a
  summation of access frequencies of mobile hosts
  in the group to the item. The calculation is done
  by the mobile host with the lowest suffix (i) of
  host identifier (Mi) in the group.

16
DCG(Dynamic Connectivity based Grouping)

• 4)  In the order of the access frequencies of
  the group, replicas of data items are allocated
  until memory space of all mobile hosts in the
  group becomes full. Here, replicas of data items
  which are held as originals by mobile hosts in
  the group are not allocated. Each replica is
  allocated at a mobile host whose access frequency
  to the data item is the highest among hosts that
  have free memory space to create it.
• 5) After allocating replicas of all kinds of
  data items, if there is still free memory space
  at mobile hosts in the group, replicas are
  allocated in the order of access frequency until
  the memory space is full. Each replica is
  allocated at a mobile host whose frequency to the
  data item is the highest among hosts that have
  free memory space to create it and do not hold
  the replica or its original. If there is no such
  mobile host, the replica is not allocated.

17
DCG(Dynamic Connectivity based Grouping)

• Data accessibility is expected to be higher since
  replicas are shared among a group of hosts
• Overhead and traffic higher than the other two
  methods since it consists more steps and needs to
  take the largest time among the three methods to
  relocate replicas in a wide range.
• the probability is higher that the network
  topology changes during executing this method,
  and in this case, the replica relocation cannot
  be done at mobile hosts over disconnected links

18
DCG example
19
Simulation Model

• 50 ? 50 flatland
• Each host randomly moves in all directions
• Movement speed is randomly determined between 0
  to d
• Radio communication range is a circle of radius
  R (1-19) fixed to 7
• Number of hosts Number of data items 40
• Each host has creates up to C replicas (1-39)
  fixed to 10
• Access frequency of each host to Di is pi given
  by one of the three cases
• Case 1 pi 0.5(1 0.01i)
• Each host has same access characteristics, access
  frequencies vary in a Small range
• Case 2 pi 0.025i
• Each host has same access characteristics,
  access frequencies vary in a Wide range
• Case 3 pi is determined as a positive value
  based on N( 0.5(10.01i) , s )
• larger the value of s, higher the difference in
  the access characteristics of the hosts
• Relocation period T (1-8192) fixed to 256
• Simulated for 59,000 time units and traffic
  measured (traffic number of hops used for
  relocating data)

20
Conclusion

• Introduced replica allocation in ad-hoc networks
  as a mechanism of improving data accessibility
• Proposed 3 replica allocation methods that use
  access patterns and the network topology
• Simulation results show that DCG gives the
  highest accessibility at the cost of increased
  traffic and SAF has the least traffic with low
  data accessibility
• The replica allocation method depends on the
  system configuration and access patterns

21
Any Questions?

• Thank you!