A Grid resource may be a digital computer or a workstation cluster. An agent at the Grid level provides Grid resources and services as a high performance computing power. Agents are the high-level abstraction of a Grid resource. Every agent consists of three main layers, from bottom to top: communication, coordination and local management layer. The local management layer acts as an agent for local grid load balancing. The coordination layer deals with the requests and organizes all the local information of the grid. The communication layer provides permit to act and deal with different agents. PACE performance prediction engine  which is a tool set for performance prediction is used by agents in Parallel and Distributed Systems.
A*: A* is a tree search technique based on an m-array tree, beginning at a root node that is a null solution . As the tree grows, intermediate nodes represent partial mappings and leaf nodes represent final mappings. Each node has a cost function, and the node with the minimum cost function is replaced by its child node. Whenever a node is added, to reduce the height of the tree, the tree is pruned by deleting the node with the largest cost function. This process is repeated until a complete mapping (a leaf node) is reached. Though the above stated heuristic algorithms have advantages, they do have their own disadvantages. OLB leads to poor makespan since it does not consider the expected execution time while mapping the meta-tasks to the machines and it is also hard to achieve dynamic load balance of jobs. MET results in severe load imbalance across the machines. Static mapping of meta-task to machine using MCT heuristic algorithm leads to poor makespan since it takes more time for a job to map to the particular machine. Max-min is appropriate only when most of the jobs arriving to the gridsystems are shortest and also Max-min outperforms Min-min. The experimental results show that Duplex, SA, GSA, and Tabu do not produce good mappings. Min-min, GA, and A* were able to deliver good performance. GA is better than Min-min by few percents, and also it has to be “seeding” the population with a Min-min chromosome to obtain its good performance. In different situations, A* produce better or worse mappings than Min- min and GA. Among the three algorithms, Min-min is the fastest algorithm, GA is much slower, and A* is very slow. Among the stated algorithms, Min-min is the simple and fastest algorithm and its good performance depends on the choice of mapping the meta-tasks to the first choice of minimum execution time. However the drawback of Min-min is that, it is unable to balance the load because it usually assigns the small task first and few larger tasks, while at the same time, several machines sit idle, which leads to poor utilization of resources. The proposed algorithm retains the advantage of Min-min algorithm and reduces the idle time of the resources, which in turn leads to better makespan.
In this paper, a new Communication Leveled DAG with Duplication CLDD algorithm is presented for heterogeneous distributed computingsystems (HeDCS). This algorithm is based on rank value to give a priority to each task. The CLDD algorithm also uses task duplication with low time complexity to minimize communication overhead. According to the simulation results, it is noted that the CLDD algorithm is better than ECTS, HEFT, CPOP algorithms in terms of time complexity, schedule length, speedup and efficiency. Performance ratio in schedule length, speedup and efficiency respectively are 17.6%, 16.9% and 16.6%. The CLDD algorithm can be tested on real applications and the development can be made on efficiency. The algorithm can be applied on cloud and gridsystems as a future work.
Cloud computing is basically seen as an evaluation. There are some roots that led to cloud computing. It is because of coming of web services and web service standards that today we can create, package and use powerful services online. So, it becomes the first root. The next and most important reason for evaluation of cloud is gridcomputing. Gridcomputing means to collect all the resources which are distributed and get easy access to them. Since different resources have different software configurations, compilers, libraries and runtime environments, there was the need of a technology that addresses these issues. Virtualization technology did the needful to enable hosting of different software applications on a single physical platform. For fair and equal access to resources, gridcomputing lacked because it considered traditional metrics’ like throughput, waiting time; which failed. Utility computing considered QoS constraints for example deadline and importance of jobs for users to compete for resources. Hardware virtualization allowed virtualization of large scale data centres and other resources (processors, memory, I/O devices) to improve sharing and maintenance. Autonomic computing was studied to improve systems by decreasing human involvement. These were the technologies that led to cloud computing. Hence, cloud
5. Earliest Deadline First: Earliest Deadline First (EDF) or Least Time to Go is a dynamic scheduling algorithm used in real-time operating systems. It places processes in a priority queue. Whenever a scheduling event occurs (task finishes, new task released, etc.) the queue will be searched for the process closest to its deadline, the found process will be the next to be scheduled for execution. EDF is an optimal scheduling algorithm on preemptive uniprocessors. 6. Minimum Execution Time: R. F. Freund et al  investigate that in contrast to OLB, Minimum Execution Time (MET) assigns each task, in arbitrary order, to the machine with the best expected execution time for that task, regardless of that machine's availability. The motivation behind MET is to give each task to its best machine. This can cause a severe load imbalance across machines.
Cloud computing is a new computing mode. It is similar to utility computing which involves a large number of computers connected through communication n/w. Cloud computing is trend to provide service as resources including hardware, software, network etc. Every service is provided over network that require high speed of network and persistence connection where its services are distributed over the network according to architecture and geo-location. It is based on pay as you go model, means it depends on matrices like usability, durability, cost, load etc. So that consumer does not need to buy any hardware, software etc. The main goal of cloud computing is to achieve higher throughput, availability, scalability, consistency guarantees, and usability, fault tolerance etc. used distributed resources . Cloud computing resources should able to solve large scale of computation problems. Cloud computing uses characteristics of Client–server model, Gridcomputing, Peer-to-peer, Mainframe computer, Utility computing to provide better services like gaming, tons of computation, message passing, network etc. Cloud computing has an advantage of delivered a flexible, very high performance, pay-as-you-go, on-demand service. Operators should guarantee to the subscribers and stick to the Service Level Agreement. Google adopts Map-Reduce scheduling mechanism schedulingalgorithms are relatively simple (First fit etc.). FIFO, default algorithm performs not so well for short jobs. Besides, Facebook proposes fair share scheduler; Yahoo raises computation ability scheduler. However, these schedulingalgorithms cannot work out a better scheduling scheme. In fact, tasks scheduling in cloud is a NP complement problem with time limit. That is to say, it is seldom impossible to search out a reasonable solution in polynomial time. To improve performance of cloud computing, efficient taskscheduling and resource management is required.
In grid, resources are distributed, heterogeneous, dynamic and instability. Yi Li et.al proposed , which is a new heuristic algorithm and based on the behavior of real ants. When the blind insects, such as ants look for food, the moving ant lays some pheromone on the ground, thus making the path it followed by a trail of this substance. While an isolated ant moves essentially at random, an ant encountering a previously laid trail can detect it and decide with high probability to follow it, thus reinforcing the trail with its own pheromone. The collective behavior that emerges means where the more the ants are following a trail, the more that trail becomes attractive for being followed. The process is thus characterized by a feedback loop, where the probability with which an ant chooses an optimum path increases with the number of ants that chose the path in the preceding steps. These observations inspired a new type of algorithm called ant algorithms or ant systems.
The motivation for providing faster exact schedulability analysis for general EDF systems is two-fold. As part of the design process many different parameter profiles may need to be checked. An automated search may even be undertaken as part of the architectural definition of the system. An efficient but accurate schedulability scheme is therefore needed. The second requirement comes from online systems. During the run-time of a system there could be new tasks arrive that need (if possible) to be added to the task set. The system must recalculate schedulability online to decide whether to allow the new tasks to enter into the system. Such online admission control gives a much higher requirement for the performance of the schedulability test as the decisions have to be made in a very short time and should not occupy too much system resource. Our scheme has the additional advantage of limiting migration costs, even in comparison to other EDF-based schemes: only up to M -1tasks, where M is the number of processors, ever migrate, and those that do, do so only between jobs. As noted in , migrations between jobs should not be a serious concern in systems where little per-task state is carried over from one job to the next.
Non-preemptive scheduling strategies [1, 7, 20, 6, 19, 16, 9] are discussed in the literature. MCP (Modified Critical Path), HEFT (Highest Level First) with estimated time algorithm , the earliest time first algorithm EFT  and dynamic level scheduling algorithm(DLS) are some well known non- preemptive schedulingalgorithms. PTS (Preemptive TaskScheduling) algorithms shows low scheduling cost and better load balance than the existing list schedulingalgorithms. Time complexities of PTS is better than time complexities of MCP, HLEFT, ETF, DLS algorithms. In these schedulingalgorithms turnaround time and CPU utilization are the metrics to evaluate the performance of the system. FCFS scheduling shows worst average job slow down than SJF . Starvation problem may be occurs in FCFS because some long jobs having more priority than short jobs may take very long execution time. This problem shows long delays and very low throughput. Job fairness parameter shows better results in non-preemptive scheduling. It is clear that non- FCFS scheduling are less fair than FCFS due to the starvation problems. Fairness of job may be calculated by the delay in time due to delayed of a later arriving job. If actual start time of a job is greater than its fair start time then the job is treated as unfair. FCFS schedulingalgorithms do not shows always better results in terms of fairness than another scheduling algorithm is that provide better response time .
GridComputing has evolved as a vital field focusing on sharing of resources. Efficient scheduling of tasks is most challenging issues in GridComputing. Load Balancing is a technique to improve utilization of resources increasing throughput managing, parallelism and to reduce the time taken for response through proper distri- bution of the jobs. Generally there are three type of phases related to Load balancing i.e. Information Collec- tion, Decision Making, Data Migration. Through this paper, we are proposing a Load balancing algorithm for optimal scheduling. It schedules the task by reducing completion time and by rescheduling waiting time of each job to obtain load balance. This algorithm works on providing optimal solution so that it minimizes the execu- tion time and expected price for the execution of all the jobs in the grid system. Load balancing algorithms are basically of two types, static and dynamic. Our algorithms in this paper based on dynamic nature load balanc- ing.
Batch Mode heuristic schedulingalgorithms (BMHA) and Online Mode heuristic algorithms. In BMHA jobs are queued and collected into a set when they arrive in the system. The scheduling algorithm will start after a fixed period of time for example First Come First Serve (FCFS), Round Robin Scheduling (RR), Min-Min and Max-Min Algorithm. In Online mode heuristic scheduling algorithm are more not efficient in cloud environment as it schedules the jobs when they arrive in the system example of this is Most Fit TaskScheduling algorithm. MFTSA in this algorithm task which fit the best in the queue are executed first. This algorithm has high failure ratio. He also explained the Resource Aware Scheduling Algorithm, a Priority based Job Scheduling Algorithm in Cloud Computing and many more then he compared all these algorithms with some fixed parameter. Saeed Parsa et al. proposed a new algorithm i.e. RASA . In this paper they took the two basic algorithms i.e. Max-Min and Min-Min. These two algorithms work on the estimated time for execution and completion of the task. In these execution time of each task is calculated on the resources. In MIN-MIN task with the minimum completion time is selected first and assigned to the resource with the minimum execution time. This procedure is followed with all the tasks. Main disadvantages of this algorithm are that in this longer tasks have to wait for the long time. But MAX-MIN algorithm selects the larger tasks first and then it selects the smaller tasks. Disadvantage of this algorithm is that in this smaller tasks have to wait for the long time.
This implementation aims towards the establishment of performance qualitative analysis on make span in VM task allocation and process according to their deadline, then implemented in CloudSim with Java language. Here major stress is given on the study of dead line based taskscheduling algorithm with heterogeneous resources of the cloud, followed by comparative survey of other algorithms in cloud computing with respect to scalability, homogeneity or heterogeneity and process scheduling. A previous study also indicates change of MIPS will affect the response time and increase in MIPS versus VM decreases the response time. When image size of VM is implemented against the VM bandwidth then no significant effect is found on response time and it remains constant for which these parameters are investigated. But in case of Cloudlet long length versus Host bandwidth a pattern is observed in which response time increases in proportionate manner. Using the modified approach the reduction in the down time of the various processes are achieved as shown in results. From the results it is clear that the proposed system used the task deadline as input parameter to improve results.
ABSTRACT: Scheduling is the process which improves the performance of parallel and distributed systems. Multiprocessor is a powerful computing for real-time applications and their high performance is purely based on parallel and distributed systems. Number of scheduling tasks in homogeneous and heterogeneous multiprocessor systems is an important problem in computing because this problem is a NP-hard problem.The execution time for individual tasks in a network are specified in a vector and refer only to the computation time. To increase the performance, reducing the processing time of the program is the main aim of scheduling. In this process we are computing rank for all nodes starting from exit node. An advanced taskscheduling in heterogeneous multiprocessor by using P-HEFT Algorithm is proposed to minimize the execution time and increase the processor utilization and load balancing for more no of tasks.
Cloud Computing has come to be perception for large scale of distributed computing and parallel processing. Cloud computing is a form of internet based computing that provides shared computer processing resources and data to computers and other devices on demand. The execution and suitability of cloud computing services always depends upon the completion of the user tasks affirmed to the cloud system. Taskscheduling is one of the main types of scheduling performed. Scheduling is the major issue in establishing cloud computing system. The schedulingalgorithms should order the jobs in a way where balance between improving the performance and quality of service and at the same time maintaining the efficiency and fairness among the jobs. This paper aims at studying various scheduling methods. A good scheduling technique also helps in proper and efficient utilization of the resources. Many scheduling techniques have been developed by the researchers like GA (Genetic Algorithm), PSO (Particle Swarm Optimization), Min-Min, Max-Min, Priority based Job Scheduling Algorithm
Grids functionally combine worldwide distributed computers and information systems for creating a universal source of computing power and information. A key characteristic of Grids is that resources are shared among numerous applications, and therefore, the amount of resources available to any given application is highly fluctuating over time. In present scenario, load balancing plays a key role. For applications that are Grid enabled, the Grid can offer a resource balancing effect by schedulinggrid jobs on machines with low utilization. A proper scheduling and efficient load balancing across the grid can lead to improved overall system performance and a lower turnaround time for individual jobs. The main objective of load balancing is to minimize the makspan time to enhance resources, utilizing parallelism, exploiting throughput managing and reduce response time through a suitable distribution of the application .
computing using Improved Max-min demonstrates achieving schedules with comparable lower makespan rather than RASA and original Max-min. The concept of load balancer aimed to distribute the tasks to different Web Servers to reduce response times was introduced in . The Web Services have gained considerable attention over the last few years. This paper presents an efficient heuristic called Ga-max-min for distributing the load among servers. Heuristics like min-min and max-min are also applied to heterogeneous server farms and the result is compared with the proposed heuristic for VOD Servers. Ga-max-min was found to provide lower makespan and higher resource utilization than the genetic algorithm. In paper  a Load Balanced Min-Min (LBMM) algorithm is proposed that reduces the makespan and increases the resource utilization. When the number of the small tasks is more than the number of the large tasks in a meta-task, the Min-Min algorithm cannot schedule tasks, appropriately, and the makespan of the system gets relatively large. Furthermore it does not provide a load balanced schedule. To overcome the limitations of Min-Min algorithm, a new taskscheduling algorithm, are proposed. It has two- phases. In the first phase the traditional Min-Min algorithm is executed and in the second phase the tasks are rescheduled so as to use unutilized resources effectively. A comparison of various load balancing algorithms is shown in Table-4.
V. Nelson, V. Uma et al16] The resources held by a single cloud are usually limited and at peak period, the organization may not be able to give the guaranteed services due to insufficient provisioning of resources. So it is essential to organize cloud systems that complement each other such as to procure resources from other participating cloud systems. However, it is difficult to provide the right resources from various cloud providers because management policies and descriptions about various resources are different in each organization. Having these differences, it is hard to provide interoperability among them. Representing cloud environment through ontology can conceptualize common attribute among the various resources semantically. Considering this fact, we propose an Inter-cloud Resource Provisioning System (IRPS) in which the resources and tasks are described semantically and stored using resource ontology and the resources are assignedusing a set of inference rules and a semantic scheduler2.
The author of this paper proposed the approach which is known as improved cost-based scheduling algorithm. The main objective of his work is to schedule groups of task in cloud computing platform, where resources are having different resource costs and different computation performance. When grouping of jobs is done, communication between jobs and resources optimizes computation/communication ratio. This algorithm measured performance of computation and cost of resources. This also increased the execution of tasks / transfer of data between tasks ratio by combining various tasks during execution. The process of combining task is usually done by after analysing the capability of different available resource and its processing. CloudSim has been used for performing the simulation and the inputs of the simulation are: average MI of tasks, granularity size of tasks, total number of tasks and task overhead time. Result of his work shows that for this particular algorithm time taken to complete tasks after grouping of tasks is very less as compared to when grouping is not done.
In centralized scheduling, a central machine that acts as are source manager to schedule jobs to all the surrounding nodes that are a part of the grid environment. It is used in situations like a computing center where resources that have the similar characteristics and usage policies. Jobs are first submitted to the central scheduler who then dispatches the jobs to the appropriate nodes those jobs that cannot be started on a node are normally stored in a central job queue. In distributed computing, multiple localized schedulers interact with each other in order to dispatch jobs to the participating nodes. There are two mechanism for a scheduler to communicate with other schedulers are direct and indirect communication. It overcomes the scalability problems which are incurred in the centralized paradigm in addition it can offer better fault tolerance and reliability. However the lack of a global scheduler which has the necessary information on available resources, usually leads to suboptimal scheduling decisions. In hierarchical scheduling, a centralized scheduler interacts with local schedulers for job submission. The centralized scheduler is a kind of a meta-scheduler that dispatches the submitted jobs to local schedulers. Similar to the centralized scheduling, hierarchical scheduling can have scalability and communication bottlenecks. However, compared with the centralized scheduling, one advantage of hierarchical scheduling is that the global scheduler and local scheduler can have different policies in scheduling jobs 
Gridcomputing technologies are increasingly being used to aggregate computing resources that are geographically distributed across different locations. Commercial networks are being used to connect these resources, and thus serve as a fundamental component of gridcomputing. Since these grid resources are connected over a shared infrastructure, it is essential to consider their effect during simulation . Simulation is the limitation of the operation of a real-world process or system over time. The act of simulating something first requires that a model be developed; this model represents the key characteristics or behaviors/functions of the selected physical or abstract system or process. The model represents the system itself, whereas the simulation represents the operation of the system over time .