Abstract: Cloud computing data centers are generally driven by services executing over cost-efficient commodity hardware. The hardware is easily scaled horizontally by the addition of more resources. However, the use of commodity hardware also results in a high failure rate of physical machines (PMs) in the cloud environment. This, in turn, causes the failure of virtual machines (VMs) provisioned on the failed physical machine. The paper proposes a connected dominating set (CDS)-based method for the construction of a virtual backbone over the data center network topology. This backbone is then utilized to develop a replication-based fault tolerance scheme to improve the reliability of VM-based services. The CDS construction approach models the network topology as a graph using the PMs as nodes. It then proposes two rules based on temperature and node degree of PMs to obtain a CDS for the topology graph. The PMs at lower CPU temperature and possessing higher node degree are selected to constitute the CDS, and, subsequently, VM replicas are placed on these CDS nodes. The performance of the proposed replication scheme is evaluated using two metric: the total bandwidth and the energy consumed during replica placement and retrieval. It is compared with a random replica placement method and an existing method of CDS construction. The simulation outcomes demonstrate that the proposed method provides better reliability and consumes fewer resources than other methods.

Key words: cloud computing, reliability, fault tolerance, connected dominating set, virtual machine, replication