A PEER TO PEER RESOURCE DISCOVERY
SCHEME FOR PROVISIONING IN CLOUD
Rayhanur Rahman
Institute of Information Technology,
University of Dhaka,
Ramna, Dhaka, Bangladesh
E-mail: rayhanur.rahman@iit.du.ac.bd
Kazi Sakib
Institute of Information Technology,
University of Dhaka,
Ramna, Dhaka, Bangladesh
E-mail: sakib@univdhaka.edu
Abstract: The ability to cater to large number of user demands without service
unavailability has become the key to the success of Cloud computing. A
fundamental challenge in building such large scale cloud based services is the
scalability and fail safe techniques for discovering and provisioning of resources
to meet expected Quality of Service (QoS). In the Cloud, existing resource
provisioning models rely on centralized information services for resource
discovery which are prone to single point failure as well as lack of scalability
and fault-tolerance abilities. In order to tackle these challenges, this work first
proposes a resource discovery scheme based on structured Peer to Peer (P2P)
architecture. As existing Distributed Hash Table (DHT) based P2P schemes
inherently do not support multi attribute data publishing and range querying
which are fundamental parts of obtaining information while making provisioning
decision, the work also addresses this inability by proposing an attribute hub
based data tuple indexing. It also proposes routing and storing data tuples
inside the nodes along with decentralized P2P oriented model for publishing
of provisioning information, retrieval and fail safe policies for node join and
leave. The work then presents a decentralized resource provisioning model using
the aforementioned resource discovery scheme and utilizing Multi Attribute
Utility Theory methods to make provisioning decisions in trade-off situations.
Simulation shows that the proposed approach is 44.24% faster on average and
45.81% faster in the case of maximum number of VMs than the centralized and
DHT based approaches respectively in case of multidimensional range querying.
The proposed system also shows less number of Service Level Objective (SLO)
violations and migrations which are about 60.27% and 83.58% respectively. Thus
the work demonstrates the effectiveness of the proposed decentralized approach
for resource discovery and provisioning in addressing scalability and resilience
to single point failure.
Keywords: Cloud Computing, Peer to Peer, Resource Provisioning, Resource
Discovery, Distributed Hash Table, Multi Attribute Utility Theory
Copyright © 2014 Inderscience Enterprises Ltd.