Вы находитесь на странице: 1из 1

Boruvkas Algorithm with Parallel Approach

Sughan Pandita*(K00365704), Pratik Patel *(K00372884), Nilesh Takale(K00364817)

Texas A & M University-Kingsville


Email: {panditasughan100, nileshtakale, pratik.ce39}@gmail.com
AbstractIn this paper, we study for
parallel approach in minimum spanning
tree which is hinge on Boruvka a sequential
algorithm. In this the target framework for
our algorithm is asynchronous. Analysis of
parallel algorithm on simple model is
similar to the LogP model. Here we want to
speed up number of processors but that will
affect the communication cost. To reduce
this cost, we can use multi core GPU and
CPU chips. In multi core CPU execution can
scale up to 8 threads and GPU
implementation is better than threads that
are running on CPU. Using this approach
cost of commutation decreases by getting
little improvement in running times.
1. INTRODUCTION
Minimum Spanning Tree (MST) of a graph
consist of set of edges that connects every
vertex in original graph. A parallel architecture
contains a group of processors that are fast and
are connected by strong communication
network. The MST has been lively for many
years with numerous of MST implementation
have been proposed and MST is often required
for efficient implementations. Using fast
parallel architecture is fast but interprocessor
communication cost is high as compared to
actual computation cost.
Graphical Processing Units (GPU) is currently
greatly used for utilizing parallelism in data
parallel algorithm. Graph algorithm is applied
on every vertex underlying execution model to
map vertex per thread. Boruvkas algorithm
provides high performance and better
scalability in multi core GPU and CPU.
Instead of using load balancing scheme [1] that
provide little improvement in running times.
We are here using GPU implementation on
multi core system which provide faster running

of parallel processor and also provide


minimum running cost.
In our proposal efficient parallel variant
implementation and design can be used to
enhance performance of the algorithm by
providing data locality and data coalescing
improved in GPU and CPU. We also provide a
efficient approach that provide contraction of
graph. Using Compressed Graph Row format
provides a newly contacted graph. This uses a
platform independent approach on both CPU
and GPU distributed system.
The rest paper consists as follow Section II
introduces information about Boruvkas
algorithm. Section III provides the related work
that directly gives the information about our
variant. Section IV presents our algorithm.
II BORUKAS ALGORITHM
In this first initialize each vertex which connect
each other by path and no vertex is connected to
vertex of another component. In this components
that are connected by the selected edges are
joined together and thus joining two components
into a new one. Union of edges at each iteration
from MST.A graph description of the algorithm
shown in Fig1.
We can provide efficient implementation by
using a disjoint set structure. The end point for
each vertex can be contacted into super-vertex
removing all the edges. Edges that can never be
part of the MST are excluded.
III. RELATED WORK

Вам также может понравиться