Ammar H. Alhusaini, Viktor K. Prasanna, and C.S. Raghavendra. A unified resource scheduling framework for heterogeneous computing environments. In Proc. HCW'99, 8th Heterogeneous Computing Workshop, pages 156--168, San Juan, Puerto Rico, April 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....requirements are unitary, while others are variant. 2. 3 Task Sequences In the theory of metacomputing, applications may be broken up into subtasks each of which may be executed on different topographical network elements, the results of which are in some way logically joined by the metacomputer [1]. Depending on the metacomputing mechanism used, the topographical structure and the location of specific elements may be more or less transparent to the end user. For the work discussed in this paper, we make the simplifying assumption that a task is an application invoked by a user, and each ....

Ammar Alhusaini, Viktor K. Prasanna, and C.S. Raghavendra. A Unified Resource Scheduling Framework for Heterogeneous Computing Environments. In Proceedings of the Heterogeneous Computing Workshop, pages 156--165, San Juan, PR, April 1999. IEEE Computer Society Press.

....and mapping of application tasks or data to those resources. To e#ectively provide these services, schedulers must evaluate the target Grid resource environment in terms of the requirements of the application itself. Many projects have successfully developed scheduling strategies for the Grid [1, 2, 12, 42, 48, 49, 50, 52, 53] While these schedulers do consider application requirements, the majority of such e#orts embed application specific details in the scheduling software itself; components that are commonly embedded include application specific performance models and strategies for mapping application data or tasks ....

....metric could be predicted execution time, resource usage cost, or throughput. In this thesis, we will consider the metric of predicted execution time as provided by an equational performance model. This is the most common application execution performance metric for run time schedulers [2, 10, 12, 8, 43, 45, 50, 53]. AART Model This model provides a structured method for specification of application resource requirements. The need to consider application resource requirements for e#ective application scheduling on the Grid is clear [8] however, since many Grid application schedulers are designed for a ....

Ammar H. Alhusaini, Victor K. Prasanna, and C.S. Raghavendra. A unified resource scheduling framework for heterogeneous computing environments. In Proceedings of the 8th Heterogeneous Computing Workshop, April 1999. 2, 8

....current loads, network conditions and topology, we concentrate on a distributed and presumably more scalable model, in which each site takes informed decisions based on its view of the Grid. Heuristics like Max min and Min Min are used for Level by Level scheduling of DAGS by Alhusaini et al. in [5]. They consider data location as a parameter but assume that resource performance characteristics are perfectly pred i ctable. What sets our work apart from other grid scheduling research is that we consider dynamic data replication as a fundamental part of the scheduling problem. 3. System Model ....

Alhusaini, A.H., Prasanna, V.K. and Raghavendra, C.S., A Unified Resource Scheduling Framework for Heterogeneous Computing Environments. in 8th Heterogeneous Computing Workshop, (1999).

....current loads, network conditions and topology, we concentrate on a distributed and presumably more scalable model, in which each site takes informed decisions based on its view of the Grid. Heuristics like Max min and Min Min are used for Level by Level scheduling of DAGS by Alhusaini et al. in [7]. They consider data location as a parameter but assume that resource performance characteristics are perfectly pred ictab le. We adapt a more realistic approach where in schedulers take decisions based on the current conditions in the Grid. Replication strategies for grids have been studied in ....

Alhusaini, A.H., Prasanna, V.K. and Raghavendra, C.S., A Unified Resource Scheduling Framework for Heterogeneous Computing Environments. in 8th Heterogeneous Computing Workshop, (1999).

....current loads, network conditions and topology, we concentrate on a distributed and presumably more scalable model, in which each site takes informed decisions based on its view of the Grid. Heuristics like Max min and Min Min are used for Level by Level scheduling of DAGS by Alhusaini et al. in [5]. They consider data location as a parameter but assume that resource performance characteristics are perfectly predictable. What sets our work apart from other grid scheduling research is that we consider dynamic data replication as a fundamental part of the scheduling problem. 3 System Model ....

Alhusaini, A.H., Prasanna, V.K. and Raghavendra, C.S., A Unified Resource Scheduling Framework for Heterogeneous Computing Environments. in 8th Heterogeneous Computing Workshop, (1999).

....when such clusters are upgraded by addition of nodes, they become heterogeneous. Thus, many organizations today have large and heterogeneous collections of networked computers. The issue of effective scheduling of tasks onto such heterogeneous clustered systems is therefore of great interest [1,11,12,18,22], and several recent research studies have addressed this problem [2,6,7,13,14,15,16,19,20,21] A simulation study comparing a number of static task matching heuristics for heterogeneous systems was reported in a recent paper [6] Randomly generated matrices were used to characterize the expected ....

....1 m row vector R is created by repeatedly generating a uniform random number in the range [1, f r ] and assigning to R[i] for 0 i m. This vector represents the relative speed of each machine and parameter f r represents the range of machine speeds, i.e. machine heterogeneity. The product of B[1] and R creates one row of the ETC matrix, i.e. ETC[1,j] B[1] R[j] Similarly, this process is repeated for each row, generating a new vector R in each step. Parameters f b and f r are varied to simulate different heterogeneous computing scenarios. The degree of task heterogeneity is varied by ....

[Article contains additional citation context not shown here]

A.H. Alhusaini, V. K. Prasanna, and C. S. Raghavendra. A Unified Resource Scheduling Framework for Heterogeneous Computing Environments, 8th Heterogeneous Computing Workshop (HCW '99), Apr. 1999.

.... locations, need to be used together [6] Mapping applications onto computational grids is a wellstudied problem [2] Most of the mapping algorithms in the literature are static and assume perfect estimation of computation and communication costs are available at compiletime (e.g. [1, 3, 5, 12, 17, 18, 20, 21]) However, at run time, computation and communication costs may differ from the estimated costs and this may greatly affect the performance of static algorithms. Several dynamic algorithms have been proposed (e.g. 10, 11, 14, 15, 16] Most static and dynamic algorithms consider compute ....

A. H. Alhusaini, V. K. Prasanna, and C. S. Raghavendra. A unified resource scheduling framework for heterogeneous computing environments. In 8th Heterogeneous Computing Workshop (HCW' 99), pages 156--165, April 1999.

....execution. Sources of input data and the execution times of the tasks on various machines along with their availability were considered simultaneously to minimize the overall completion time. We have developed several heuristic algorithms to solve the above problem. These results are published in [1]. Although we considered multiple resource requirements, tasks were not required to access different types of resources simultaneously. In the second problem, we considered mapping a set of applications in a heterogeneous computing (HC) system where application tasks require concurrent access to ....

Ammar Alhusaini, Viktor Prasanna, and C. S. Raghavendra. A Unified Resource Scheduling Framework for Heterogeneous Computing Environments. In Proceedings of the Eighth Heterogeneous Computing Workshop (HCW '99), pages 156--165, San Juan, PR, April 1999.

.... Additionally, MSHN team members have performed extensive research into accounting for dependencies between applications or processes that make up a single application [40] 47] 48] 54] This includes promising results from investigating data dependencies and mapping of iterative applications [1][4] 5] 6] 11] 3.2. Resource Status Server and Resource Requirements Database research issues Part of the MSHN team s investigation has been aimed at determining what information should be stored in the Resource Requirements Database and maintained by the Resource Status Server. First, a ....

A. H. Alhusaini, V. K. Prasanna, and C. S. Raghavendra, "A unified resource scheduling framework for heterogeneous computing environments," Proc. 8 th IEEE Heterogeneous Computing Workshop, April 1999.

No context found.

Ammar H. Alhusaini, Viktor K. Prasanna, and C.S. Raghavendra. A unified resource scheduling framework for heterogeneous computing environments. In Proc. HCW'99, 8th Heterogeneous Computing Workshop, pages 156--168, San Juan, Puerto Rico, April 1999.

No context found.

A. H. Alhusaini, V. K. Prasanna, C.S. Raghavendra, "A Unified Resource Scheduling Framework for Heterogeneous Computing Environments", Proceedings of the 8th IEEE Heterogeneous Computing Workshop, Puerto Rico, 1999, pp. 156 - 166.

No context found.

A. Alhusaini, V. Prasanna, and C. Raghavendra. A unified resource scheduling framework for heterogeneous computing environments. In 8th Heterogeneous Computing Workshop (HCW), pages 156--165, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC