Today, cloud computing serves as a request response model, where a client makes request for various available services on "pay as you go basis". Cloud computing offers a dynamic flexible resource allocation phenomenon. For reliable and guaranteed services there must be a scheduling mechanism that all resources are efficiently allocated to satisfy the customer's request. Cloud services are based on scalability, availability, security and fault tolerance features. Service provisioning in cloud is based on SLA. Service level agreement is the terms of cloud provider's contracts with customers to define the level(s) of service being sold in plain language terms. QoS (quality of service) plays important role in cloud environment. Resource scheduling and service deployment is done by considering multiple SLA parameters like CPU requirement, network bandwidth, memory and storage. In this paper we propose an algorithm which perform resource preemption from low priority task to high priority task and advanced reservation for resources considering multiple SLA parameters for deploying service. This algorithm is also effective for fault tolerance mechanism.
Cloud computing is "providing computing resources and applications over the internet" using pay as you go model. It is also known as internet computing, here a pool of resources such as memory, processor, network and bandwidth, virtually distributed across the internet. If a customer wants to use the services of cloud provider then it has to pay cost according to services using real time as per requirements. Cloud computing provides globalize sharing of resources and unlimited storage capacities.
As we know that any number of customers can make requests to cloud provider, if SLA based agreement takes place that means cloud provider are able to attain the corresponding request from the user, this is done by efficiently scheduling of resources and deploying applications on proper VMs. Resource scheduling means multiplexing of several user requests on same physical structure. In this time there exists more work already done on scheduling of resources in clouds this approach is based on global resource deployment by considering one SLA objective such as cost of execution, time of execution, minimum resources etc.
When a request is submitted by the client, it's firstly partitioned into several subtasks. Now there are four main responsibility of cloud scheduler, 1) find appropriate manner or order to execute a task, 2) find appropriate resource allocation schedule for task, 3) the scheduler should be fault tolerance to schedule overheads and terminate the task, 4) find a manner for migration of tasks in an effective manner. Using the resource allocation phenomenon these problems can be solve.
The scheduling can be two types, global scheduling and local scheduling. Local scheduling is a type of scheduling where localize resources are use to satisfy the user's request within one single cloud. Global scheduling is where all the resources from multiple clouds are treated as one single unit to complete the user's request. Typically efficient provisioning required two distinct processes, 1) initial static planning - local scheduling where all the VMs are mapped to physical resources. 2) Dynamic resource provisioningcreate new VMs, migration on VMs, dynamically response according to workload. Here step 1 is set up stage, it is generally performed at the time of set up a cloud, and when maintenance is done by the source. Whether step 2 run repeatedly at the allocation time. There are various challenges arise in this area for researchers such as scalability, multitenency, security, dynamic resource allocation and fault tolerance.
In this paper we focus on dynamic resource provisioning, we present a scheduling heuristic considering multiple SLA objectives, such as amount required CPU, network bandwidth, and cost for deploying applications in clouds. The scheduling present a flexible on demand resource allocation strategy included advanced reservation and preempt-able mechanism for resources. Our proposed algorithm dynamically responds to requested resource for the task. First it's locally checks for the availability of resource; if resource is free then it deployed new VMs to current task, If resource is not available then it's create new VM from globally available resource; if global resources are not available then it will check for resource if it's preemptable then it's migrate processes otherwise put the task into waiting list and apply advanced reservation scheme.
Organization of papers is as follows: In section 2 we discuss research works related to this topic, in section 3 models for resource allocation and task scheduling in [aas cloud computing system and consist the previous algorithms that is being used in our algorithm, section 4 discusses the proposed method, section 5 shows the simulation results, section 6 conclude the paper and future work.
II. RELATED WORK
Jiayin Li  presents a resource optimization mechanism in heterogeneous IaaS federate multi cloud systems, that enable preemptable task scheduling with resource allotment model, cloud system model, local mapping and energy consumption, and application model. It is suitable for autonomic future in cloud and VMs. They proposed online dynamic algorithms for resource allocation and task scheduling. [n proposed cloud resource phenomenal every data center has a manager server, the communication and resource allotment scheme works between various servers of each data center for share workloads among multiple data servers. The workload sharing makes a large resource pool of flexible and cheaper resources to resource allocation.