Design of cloud middleware framework for resource management

Abstract

Cloud computing, an emerging technology, has gained momentum exponentially in the recent years as many of the IT services of the organization are deployed in cloud. Several cloud providers offer same or different cloud services to the users but with different pricing policies, performance attributes and varying features. Such knowledge on the distinct offers specified by the providers, their capabilities, prospective SLAs are not easily accessible by the consumers due to the lack of a generic middleware model. Hence it becomes challenging to provide the best match between the demand of consumers and resources of cloud. From consumer s perspective, identifying, mapping and selecting of suitable resource for each tasks remains an open challenge. From provider s perspective, it is difficult to publicize their service competencies, to acquire maximum users to avail the services and to efficiently manage the resources as per demand. As the resources provided by different providers vary in configuration it is difficult to manage the heterogeneity and transparency in resource management of cloud. Therefore, it is imperative to design a cloud middleware framework to enable the users to identify the appropriate services and act as an intermediary to enable the providers to manage the compute and storage resources. Therefore, a new paradigm shift in cloud is the design of a Cloud Middleware Framework model, namely CMF model, which caters to manage the resources like compute servers, storage and service and is made available for the stakeholders. Resource management comprises of resource allocation, resource mapping, resource provisioning and resource adaptation that directly affects the performance of cloud. Precise and accurate allocation is required to maximize the usage of resources. Existing system considered tasks based on the deadline and categorize as high, medium or low priority ones and allocate resources arbitrarily. However it fails to address the task type and to identify the relationship between tasks and the resource demands. Therefore, a novel Task aware Autonomic Resource Management using Neural Network (TARNN) framework is designed that can classify the tasks based on their parameters, its type i.e., if it is CPU, network, I/O or storage intensive ones, identify the relationship with the resources and allocate suitably. However, to identify and improve the acquaintance of the tasks with the resources in the context of scheduling, a modified Particle Swarm Optimization (m-PSO) algorithm is proposed to schedule the tasks to appropriate resources based on resource demands. Resources are hugely available in the resource pool and therefore searching the entire set will increase the resource-task mapping time during allocation. Therefore, in order to reduce the search space and minimize the scheduling time, a hybrid resource provisioning algorithm is devised that selects only a subset of resources from a large pool of resources for scheduling. newline