The CMs is basically Agent based Scheduler, hence it is implemented in two parts which are named as Megh and Chatak.
|
The CMs is basically Agent based Scheduler, hence it is implemented in two parts which are named as Megh and Chatak. |
One of the objectives of computational grids is to offer applications the collective computational power of distributed but typically shared heterogeneous resources. Unfortunately, efficiently harnessing the performance potential of such systems (i.e. how and where applications should execute on the grid) is a challenging endeavor due principally to the distributed, shared and heterogeneous nature of the resources involved. This project tries to propose a tool to aid the design and evaluation of scheduling policies suitable for efficient execution of parallel applications on computational grids.
The objective of computational grids is to coordinate shared, distributed heterogeneous resources to work as a single computational resource. The availability of such an environment opens new horizons for research in areas previously unexplored or limited for economic or impractical reasons. The existence of long distance, low cost, high performance networks is encouraging the development of applications which take advantage of geographically dispersed resources. However, relatively few applications exploit the computational power available from such environments efficiently. Due to the diversity of resources, their dynamic behavior and the instability generally encountered in grids, developing applications capable of executing efficiently in such environments is still a challenge. Users of parallel system frequently complain of the difficulty of achieving a reasonable fraction of the theoretical peak performance from the systems they use. In environments like grids with so much variation, it is extremely difficult for users (typically scientist and engineers with little system knowledge) to decide, for each application and for each execution, which resources would be the most appropriate. The challenge is to execute applications efficiently and robustly in grid environments, without placing this burden on the programmer or the user.
While there have been several proposals of high performance global computing systems, scheduling schemes for the systems have not been well investigated. The reason is difficulties of evaluation by large-scale benchmarks with reproducible results. This Chatak Megh Samvad (CMs hereafter) performance evaluation system would allow analysis and comparison of various scheduling schemes on a typical high-performance global computing setting. CMs can simulate various behaviors of global computing systems, especially the behavior of networks and resource scheduling algorithms. The behavior of the network and the resource Scheduling Algorithms is faithfully carried out by the Condor HTC software. The scheduling algorithm used is the UP DOWN Scheduling Algorithm for the resource scheduling.
High performance global computing systems fueled by the rapid progress of high-speed networks and computing resources are now regarded as the computing platform of the future. In order to effectively employ computing resources therein, most proposed global computing systems embody a resource scheduling framework whose components monitor the global computing environment and predict availability of the resources. For effective investigation and objective comparison of scheduling algorithms and the implementation of the scheduling frameworks, large-scale benchmarks with reproducible results under various environments parameterized by the following constituents over time are required:
Servers --- architecture, performance, load and variance.
These parameters have been included in the scheduling scheme, to be detailed later in the architecture of the CMs project. These factors of machines influence the scheduling decisions made by the CMs scheduler.
However,
reproducibility over a wide-area network is extremely costly to
achieve, if not impossible. Thus, currently it is unrealistic to
compare the different scheduling
algorithms proposed
by other researchers, let alone compare the systems themselves. Cost
and scale of possible benchmarks are also extremely limited. The
resulting lack
of impartial comparative approaches is a great hindrance to global computing research and deployment. In order to resolve this situation, we are building a performance evaluation system that would allow analysis and comparison of various global computing systems under reproducible, controlled environments, called CMs. The current version of CMs mainly focuses on the evaluation of different scheduling algorithms and schemes based on a canonical (orthodox) model of high-throughput global computing system. The CMs Project tries to emulate the Grid as far as it can and give the researcher the results of the tests reliably aping the way scheduling scheme would work in a real world grid computing environment. The main aim is to prove that a scheduling scheme can be implemented in such a way, trying to divert the job level scheduling to a underlying software such as Condor and retains the application level scheduling.