Concurrent Online Tracking of Mobile Users

1
Concurrent Online Tracking of Mobile Users

• Baruch Awerbuch, David Peleg
• Interpreted by Ju Wang
• CSAG-UCSD

2
Outline

• Problem description
• Regional matching
• High level view of the scheme
• How this scheme works example
• Discussion

3
Problem Description

• Assumptions
• A network of hosts with fixed topology.
• Users are moving inside the network from one host
  to another.
• The problem
• Construct a distributed directory to keep track
  of user location.
• Two operations Move and Find

4
Problem Description (cont.)

• Challenges
• Minimize the cost for both Find and Move
  operations, which are conflicting.
• Concurrency issues
• Try to locate a moving user.
• This paper tried to solve it, but the solution is
  not really satisfying.
• We first ignore the concurrency issue, and come
  back to it in discussion section.

5
Theory Behind Regional Matching

• Formal definition
• Given a set of vertices V v1, , vn
• Given another set D d1, , dm
• For each vertex vi?V, define two sets R(vi) and
  W(vi), where R(vi)?D and W(vi)?D.
• Regional match
• For ?x,?y?V, R(x) ? W(y) ? ?

6
Theory Behind Regional Matching (cont.)
A
Write Set for host y B, C
B
Read Set for host y A, D
y
C
D
x
Host
Write Set for host x A, C
Directory Server
Read Set for host x B, D
7
High-Level View
Kth Region (Graph G)
Diameter of G, D(G)

• D(G) ? 2k (Kth region has a diameter of size 2k)
• A user v on any host in G should be able to
  locate another user u on any host in G using
  regional matching.

8
High-Level View
(K-1)-level regional directory servers

• It is not a partition of the network.
• Any two hosts within distance of 2k-1 has the
  regional mapping property, namely , R(x) ? W(y) ?
  ?

9
High-Level View
(K-2)-level regional directory servers
10
High-Level View

• For each user u, the system maintains two
  vectors
• A(u) ltR_Addr1(u), , R_Addrk(u)gt
• Caches addresses in the ith regional directory
• C(u) ltC1(u),, Ck(u)gt
• Counter to remember how far u has traveled since
  last update in the ith regional directory
• Both vectors are kept on the host where user
  resides, and will migrate with the user.

11
High-Level View
B
A

• The information in regional directory might be
  out-of-date. Use forward pointers to fix this
  problem.

12
Find Operation
13
Move Operation
A(u) ltx4, x4, x3, x2, x2, x1gt
C(u) lt0, 0, 3, 5, 5, 25gt
x1
x2
x3
x4
20
2
3
14
Move Operation
C(u) lt0, 0, 3, 5, 5, 25gt
x1
x2
x3
x4
20
2
3
1
C(u) lt1, 1, 4, 6, 6, 26gt
x5
A(u) ltx4, x4, x3, x2, x2, x1gt
Remove entry in W1(x4), W2(x4), W3(x3)
A(u) ltx5, x5, x5, x2, x2, x1gt
Add entry in W1(x5), W2(x5), W3(x5)
C(u) lt0, 0, 0, 6, 6, 26gt
Remove forward pointer in x3 Add forward pointer
in x2
15
Discussion

• Concurrency issue
• Their solution
• Regionally restrict migration
• Maybe too restrictive?
• Other solutions?
• Retry? Allow failure?
• ?
• Advantage of this scheme
• True distributed scheme
• Better use of locality than partitioning the
  network.

16
Discussion (cont.)

• Restrictions
• Assumes static network topology
• How easy it is to support dynamic network
  topology is unclear in the paper. Need more
  knowledge about the math regional matching
  problem.
• Is this scheme vulnerable to failures or attacks?
• Can be extended to tolerate failures or attacks
  by adding redundancy in the matching.
• For example, we can require that for any x, y in
  a region, W(x) and W(y) must have at least n
  common elements. (n-redundancy)