Madhulina Sarkar

Resource requirement prediction using clone detection technique

In order to maintain the QoS requirements of jobs running on a large distributed system, like Cloud and Grid environments, resource requirements of jobs should be predicted prior to their submission, and on the basis of this prediction, appropriate resources can be selected for their execution. However, because of the dynamic and heterogeneous nature of the modern distributed systems, estimation of resource requirements is a challenging task. This paper presents a feedback-based job modeling scheme based on clone detection technique. In this scheme, the execution data for each job which runs in the environment is stored in Execution History. A newly submitted job is analyzed to find its clones from the execution history and on the basis of the data stored in the execution history, the resource requirement of the new job is predicted. Different levels of clones are discussed in this paper and a metric-based clone detection technique is presented. An automatic resource requirement prediction scheme for jobs is proposed. The paper also evaluates a preliminary implementation of the scheme and discusses the results of using the scheme for some test codes.

Implementation of a Resource Broker for Efficient Resource Management in Grid Environment

A resource broker is an essential component in a Grid environment that can assist in the selection of a right resource provider for a job in all aspects. In this paper we present the design and implementation of different phases of resource brokering strategies within a multi- agent framework. These strategies help in finding out an optimal usage of resources for executing multiple concurrent jobs in a Grid environment. The paper also presents preliminary results and demonstrates the effectiveness of our strategies.

Feedback-Guided Analysis for Resource Requirements in Large Distributed System

Resource management is one of the focus areas of Grid which identifies Job Modeling to be a very important part of it. A proper Job Modeling can be helpful in allocating jobs to their most suitable resource providers in Grid. This paper presents a feedback-guided Automatic Job Modeling technique that describes the process required to identify the most suitable resource provider for a particular job.

A Hybrid Clone Detection Technique for Estimation of Resource Requirements of a Job

Resource requirement estimation in large distributed systems is a difficult job because of the heterogeneity and dynamism of the environment involving modern distributed systems. A feedback-based job modeling scheme based on clone detection technique was proposed in [6]. This paper extends the taxonomy of clones proposed by other researchers [1] in order to make resource requirement prediction more effective. It also presents a hybrid clone-detection technique, consisting of metrics-based, PDG-based and AST-based clone detection, to make the clone detection process more reliable and robust.

Optimizing resource allocation for multiple concurrent jobs in grid environment

In a dynamic environment like grid, it is difficult to manage resources at application or user level. In order to support application execution in the context of grid, a resource broker is essential and the task of resource brokering in such a heterogeneous, fast changing, distributed environment is non-trivial. In this paper we present the design and implementation of resource brokering strategies within a multi-agent framework. These strategies help in finding out an optimal allocation of resources for executing multiple concurrent jobs in a grid environment. We discuss the different stages in resource brokering and their implementation within the framework. The paper also presents results of a preliminary implementation and demonstrates the effectiveness of our strategies.

Implementation of execution history in non-relational databases for feedback-guided job modeling

A feedback-guided Resource Requirement Prediction technique has been described in [2]. An Execution History is built and maintained for all the jobs which are executed in a large distributed system like Grid. When a new job arrives, its clones are sought for in the Execution History and if clones are found, relevant performance information are retrieved and used for estimating resource requirements. In this paper, we focus on the implementation details of Execution History. Instead of using relational databases, here we have used NoSQL database, MongoDB. The reasons for using MongoDB are discussed. Techniques for storing and retrieving data from the Execution History are also described. Overheads for such operations are measured and presented in the result part of this paper.

Prediction of resource requirement using feedback on job execution performance

Abstract One of the important issues for proper usage of Grid is selection of suitable resources for jobs. Precise estimation of resource requirements for jobs is important in order to ensure efficient use of Grid resources. This paper gives an overview of an efficient job modeling technique for allocation of jobs onto the resource providers in Grid. The proposed job modeling depends on the feedback gathered from the previous executions of different jobs. The paper mainly focuses on the technical implementation details of collection of hardware performance monitoring data using the PAPI tool. The performance monitoring data are later used as feedback while analyzing the resource requirements of the job.

A Heuristic-Based Resource Allocation Approach for Parallel Execution of Interacting Tasks

Heterogeneity and complexity of distributed computing increases rapidly as high speed processors are widely available. In modern computing environment, resources are dynamic, heterogeneous, geographically spread over different computational domains and connected through different capacity of high speed communication links. In a large distributed environment a modular program can be considered as a set of loosely coupled interacting modules/tasks (since all the modules/tasks are considered as simultaneously and independently executable) and represented by task interaction graph (TIG) model. Parallel execution of these interacting modules/tasks is highly preferred to reduce the overall completion time of a program. During parallel execution of tasks, the communication overhead due to message passing may increase the cost of parallel execution. Parallel execution of tasks is chosen if and only if parallel execution cost together with communication overhead is less than serial execution cost. So, resources are to be allocated such that advantage of parallel execution is maintained. In this paper, for any task and resource graph, we propose a heuristics based approach to find out an optimal number of tasks that can be executed in parallel on a set of resources where they can be executed.