Observation%20on%20Parallel%20Computation%20of%20Transitive%20and%20Max-closure%20Problems

1
Observation on Parallel Computation of Transitive
and Max-closure Problems
2
Motivation

• TC problem has numerous applications in many
  areas of computer science.
• Lack of course-grained algorithms for distributed
  environments with slow communication.
• Decreasing the number of dependences in a
  solution could improve a performance of the
  algorithm.

3
What is transitive closure?
GENERIC TRANSITIVE CLOSURE PROBLEM (TC) Input a
matrix A with elements from a semiring S lt ?,?
gt Output the matrix A, A(i,j) is the sum of
all simple paths from i to j lt ? , ? gt
TC lt or , and gt boolean closure - TC of a
directed graph lt MIN, gt all pairs shortest
path ltMIN, MAXgt minimum spanning tree all(i,j)
A(i,j)A(i,j)
4
Finegrain and Coarse-grained algorithms for TC
problem

• Warshall algorithm (1 stage)
• Leighton algorithm (2 stages)
• Guibas-Kung-Thompson (GKT) algorithm (2 or 3
  stages)
• Partial Warshall algorithm (2 stages)

5
Warshall algorithm
k
k1
k2

• for k1 to n
• for all 1?i,j?n parallel do
• Operation(i, k, j)
• ----------------------------------
• Operation(i, k, j) a(i,j)a(i,j) ? a(i,k) ?
  a(k,j)
• ----------------------------------

k
k1
k2
Warshall algorithm
6
(No Transcript)
7
Coarse-Grained computations
A11
A24
n
A32
n
8
Naïve Course Grained Algorithms
9
II
I
10
Course-grained Warshall algorithm

• Algorithm Blocks-Warshall
• for k 1 to N do
• A(K,K)A(K,K)
• for all 1 ? I,J ? N, I ? K ? J
  parallel do
• Block-Operation(K,K,J) and
  Block-Operation(I,K,K)
• for all 1 ? I,J ? N parallel do
• Block-Operation(I,K,J)
• --------------------------------------------------
  --------------------
• Block-Operation(I, K, J) A(I,J)A(I,J) ? A(I,K)
  ? A(K,K) ? A(K,J)
• --------------------------------------------------
  --------------------

11
Implementation of Warshall TC Algorithm
k
k
k
k
k
The implementation in terms of multiplication of
submatrices
12
II
I
13
Decomposition properties

• In order to package elementary operations into
  computationally independent groups we consider
  the following decomposition properties
• A min-path from i to j is a path whose
  intermediate nodes have numbers smaller than min
  (i,j)
• A max-path from i to j is a path whose
  intermediate nodes have numbers smaller than
  max(i,j)

14
(No Transcript)
15
KGT algorithm
16
An example graph
17
What is Max-closure problem?

• Max-closure problem is a problem of computing all
  max-paths in a graph
• Max-closure is a main ingredient of the TC
  closure

18
Max-Closure --gt TC

• Max-to-Transitive
• performs 1/3 of the total operations

Max-closure algorithm

• Max-closure computation performs 2/3 of total
  operations

The algorithm Max-to-Transitive reduces TC to
matrix multiplication once the Max-Closure is
computed
19
A Fine Grained Parallel Algorithm
Algorithm Max-Closure for k 1 to n do for all
1 ? i,j ? n, max(i,j) gt k, i?j parallel do
Operation(i,k,j)

• Algorithm Max-to-Transitive
• Input matrix A, such that Amax A
• Output transitive closure of A
• For all k ? n parallel do
• For all i,j max(i,j) ltk, i?j
• Parallel do Operation(i,k,j)

20
Coarse-grained Max-closure Algorithm

• Algorithm CG-Max-Closure Partial
  Blocks-Warshall
• for K1 to N do
• A(K,K) A(K,K)
• for all 1 ? I,J ? N, I ? K ? J parallel do
• Block-Operation(K,K,J) and
  Block-Operation(I,K,K)
• for all 1 ? I,J? N, max(I,J) gt K ? MIN(I,J)
  parallel do
• Block-Operation(I,K,J)
• --------------------------------------------------
  ------------------------------
• Blocks-Operation(I, K, J) A(I,J)A(I,J) ?
  A(I,K) ? A(K,J)

21
Implementation of Max-ClosureAlgorithm
k
k
k
k
k
The implementation in terms of multiplication of
submatrices
22
Experimental results
3.5 h
23
Increase / Decrease of overall time

• While computation time decreases when adding
  processes the communication time increase
• gt there is an ideal number of processors
• All experiments were carried out on cluster of 20
  workstations
• gt some processes were running more than one
  worker-process.

24
Conclusion

• The major advantage of the algorithm is the
  reduction of communication cost at the expense of
  small communication cost
• This fact makes algorithm useful for systems with
  slow communication