Performance Improvement of Cloud Computing Environment through Efficient Consolidation of Virtual Machines

Introduction:

Cloud computing is emerging as a commodious resource of computational power in recent years. It provides elastic and scalable on-demand resources for efficient execution of different applications. The ability to scale up and down the computing platform according to the application requirements and user’s budget, makes the Cloud a cost-effective and timely solution to build HPC platform (e.g., clusters, Grids, etc.) especially to the needs of HPC users. One of the key features characterizing Cloud computing is virtualization. Virtualization technology allows better flexibility and customization to HPC applications (e.g., compute intensive applications) by minimizing the number of consolidated resources in shared resource pools. By this technology, the Cloud providers create multiple VMs (Virtual Machines) instances on a single physical server. Therefore, users are able to provision resources on-demand on a pay-as-you-go basis. The results from past research on virtualized Cloud data centers have outlined major limitations (e.g., QoS) which have been resulted from insufficient network performance, resource heterogeneity and multi-tenancy and etc., for running HPC applications on Cloud. However, Cloud advanced in recent years and now is going to solve some of these problems by leading to heterogeneous configurations in processors, memory, and network. They have already been adopted the VM placement strategies in different scenarios such as intensive and business applications in which, the user can scale up and down re-sources when required and finally drop them out when the task is done. Therefore, HPC community has discovered Cloud computing facilities as a potential target system. This is one of the main reason motivating many users and organizations to port HPC applications to Cloud.

HPC applications suffer from network latency and high overheads under virtualized resources in Cloud environment. This is due to the variability of multiple workloads and large-scale distributed platforms and computing elements. To overcome this problem, we need an efficient dynamic VM consolidation and resource allocation strategy to utilize the consolidation of shared resources. Therefore, it is necessary to take into account the network topology and bandwidth to accurately evaluate applications involving communicating task and VMs allocation.

Our recent achievement:

In our last research work [1], we have enhanced minimum migration time policy by means of simulation for compute intensive applications such as MPI on Cloud data centers in order to improve QoS (Quality of Service) and also to reduce the number of VM migrations. We used simulation-based approach because it offers significant benefits to test performance of provisioned services and polices in a controllable environment. To meet these requirements, we used CloudSim which is an extensible simulation toolkit for modeling and simulation of Cloud computing environments. Our results emphasize the validity of our method by conducting significant improvement in QoS and VM migration through combination of our proposed VM selection policy with most recent existing overloading detection and placement policies.

Proposed project:

One of the recent challenges of Cloud features is that, virtual machines migration of HPC applications creates significant network overhead and the data latency due to over subscription of network resources which will affect QoS offered to the customers. Therefore, it is necessary to take into account the network topology and band-width to accurately evaluate applications involving communicating task and VMs allocation. So, we plan to extend our previous work [1] to model comprehensive Cloud computing environment for different application and service models by developing accurate scheduling and resource management mechanisms. We need to capture message routing and latency behavior of intensive (work loaded) HPC applications caused by VMs migration in order to tune the performance bottlenecks. This is achieved by accurate provisioning of resource allocation under varying load, system size and user configuration. Then we will be able to reduce the energy consumption and to improve QoS. Later, this would be a small step towards Green Cloud computing, if we could then employ our development on real Cloud environment.

References:

[1] Rashid Hassani, Shiv. R. P. N Amgoth, Peter Luksch, “Efficient Consolidation of Virtual Machines for HPC Applications in Cloud”, International Journal of Intelligent Information Processing (IJIIP), ISSN: 2233-9426, vol. 5, no. 4, pp. 19-26, Sep 2015.

Contact for more details:

Rashid Hassani
E-Mail: rashid.hassani(at)uni-rostock.de