R. Freund, T. Kidd, D. Hensgen, and L. Moore. SmartNet: A scheduling framework for heterogeneous computing. In Proceedings of the International Symposium on Parallel Architectures, Algorithms, and Networks (ISPAN-96), pages 514--521, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....computational complexity of the algorithm; the appropriate architecture specific constants must also be determined. To our knowledge, this is the first system to use toolspecific, run time inputs to predict the resource usage characteristics of runs. Other work on resource usage prediction (e.g. [3, 4, 9, 10, 11]) utilizes tool specific analytical expressions or statistical data obtained from past runs (e.g. average execution time) to predict future resource usage. Results show that such heuristics can be used to identify better schedules. These approaches were not used here because the resource usage ....

R. Freund, T. Kidd, D. Hensgen, and L. Moore. SmartNet: A scheduling framework for heterogeneous computing. In Proceedings of the International Symposium on Parallel Architectures, Algorithms, and Networks (ISPAN-96), pages 514--521, 1996.

....Quality of Service (QoS) requirements. Task scheduling in a distributed system is a classic problem (for a detailed classification see [5, 6] Recently, there have been several works on scheduling tasks in metacomputing systems. Scheduling independent jobs (meta tasks) has been considered in [2, 11, 14]. For application tasks represented by Directed Acyclic Graphs (DAGs) many dynamic scheduling algorithms have been devised. These include the Hybrid Remapper [20] the Generational algorithm [9] as well as others [15, 18] Several static algorithms for scheduling DAGs in metacomputing systems ....

....subtask. 2.3. Problem Statement Our goal is to minimize the overall execution time for a collection of applications that compete for system resources. This strategy (i.e. optimizing the performance of a collection of tasks as opposed to that of a single application) has been taken by SmartNet [11] and MSHN [22] On the other hand, the emphasis in other projects such as AppLes [3] is to optimize the performance of an individual application rather than to cooperate with other applications sharing the resources. Since multiple users share the resources, optimizing the performance of an ....

[Article contains additional citation context not shown here]

R. Freund, T. Kidd, D. Hensgen, and L. Moore, "SmartNet: a scheduling framework for heterogeneous computing," The International Symposium on Parallel Architectures, Algorithms, and Networks, Beijing, China, June 1996.

....in emulating an ideal user in terms of its resource management and usage policies. 1. Introduction It is now recognized that the heterogeneous nature of the network computing environment cannot be effectively exploited without some form of adaptive or demand driven resource management (e.g. [10, 11, 12, 14, 18, 27]) A demand driven resource management system can be characterized by its ability to make automatic cost performance tradeoff decisions at run time. Such decisions require that the infrastructure be able to decide how (which implementation e.g. sequential versus parallel) and where (which ....

....which such expressions can be determined with relative ease (e.g. as with numerical software) As a result, this approach is of limited use in a computational grid environment. Some systems address this limitation by using analytical expressions in conjunction with other approaches SmartNet [11, 12], for example, estimates resourceusage characteristics by employing a combination of statistical analysis and optional analytical expressions provided by users. The design objectives of an application performance modeling system, as stated by Berman [10] are: 1) to utilize dynamic information to ....

R. Freund, T. Kidd, D. Hensgen, and L. Moore. SmartNet: A scheduling framework for heterogeneous computing. In Proceedings of the International Symposium on Parallel Architectures, Algorithms, and Networks (ISPAN-96), pages 514--521, 1996.

....11] etc. Other systems have been developed to allow workload rebalancing to occur while an application is running [10, 4] These systems focus primarily on redistribution to utilize idle workstation cycles, while our scheme focuses on adaptation to dynamic loads in a shared environment. SmartNet[7] accounts for system heterogeneity and resource sharing when making performance estimates and scheduling decisions, but does not support redistribution in response to changing resource loads. The PARFORM[5] Dome [1] and MARS [8] projects have investigated the use of adaptive scheduling for ....

R. Freund, T. Kidd and L. Moore, SmartNet: a scheduling framework for heterogeneous computing, Proc. 2nd Int. Symp. Parallel Architectures, Algorithms, and Networks, (1996), pp. 514-521.

....to use the choice of time elapsed divided by the expected execution time to determine what percentage of the previously computed total normalized useful work has been done. For example, the VHM scheduling system upon which the work based methods were originally installed and evaluated, SmartNet (Freund et al. 1996), provides estimated completion times for the tasks it schedules. 10 excavation work is normalized based on the amount of work required to remove one cubic yard of dirt. This normalization of the work greatly reduces the complexity involved in calculating excavation costs. Thus, the question ....

....called Speed O has been constructed at the Naval Command, Control, and Ocean Surveillance Center (NCCOSC) in San Diego according to the principles described in Section 2. It is being used to measure the performance of VHMs scheduled by SmartNet, a heterogeneous scheduling tool developed at NCCOSC (Freund et al. 1996). Performance measurement by Speed O may be made in real time as well as a replay mode. Speed O consists of two components: computational and graphical. 3.2 Computational Component Speed O tracks the state of tasks being performed in the VHM such that its current and U work S current S ....

[Article contains additional citation context not shown here]

Freund, R., Kidd, T., Hensgen, D., Moore, L., "SmartNet: A Scheduling Framework for Heterogeneous Computing," Proceedings of 2nd International Symposium on Parallel Architectures, Algorithms, and Networks, Beijing, China, June 1996, pp. 514-521.

....research in all other aspects of distributed object computation that will be needed in future versions of such standards in order to first isolate, then solve some of the difficult resource management problems. 1.1. Background MSHN evolved in part from a scheduling framework called SmartNet [19][29]. SmartNet s goal was to be able to wisely schedule sets of compute intensive jobs, some of which may require the execution of multiple processes, onto members of a suite of heterogeneous computers. SmartNet provides a sophisticated scheduling module that had been successfully integrated with many ....

T. Kidd, D. Hensgen, R. Freund, and L. Moore, "SmartNet: a scheduling framework for heterogeneous computing," Proc. 2 nd Intl. Symposium on Parallel Architectures, Algorithms, and Networks, June 1996, pp. 514-521.

.... The Problem This paper is the first in a series of papers about the software architecture for our Management System for Heterogeneous Networks(MSHN) MSHN leverages our experience in designing, building, and experimenting with SmartNet, which is a scheduling framework for heterogeneous computing [7]. MSHN is an extensible software architecture that is layered on top of native operating systems and network protocols. MSHN is responsible for determining which large scale resources, such as compute servers and file servers, should be used to execute particular requests, and shepherding the ....

Kidd, T., Hensgen, D., Freund, R., and Moore, L. Smartnet: A Scheduling Framework for Heterogeneous Computing. Proceedings of the IEEE International Symposium on Parallel Architectures, Algorithms, and Networks (ISPAN '96) (June 1996), 514--521.

....will be set to some very large number. As we describe these algorithms we will consider these expected run times as elements of a 2 dimensional, n by m matrix called A. That is, A[i; j] is the expected run time of job i on machine j. The O(nm) algorithm, which, like in the SmartNet documentation [6], we will call Fast Greedy, considers the jobs in the order requested 1 . It first determines the value A 1;j , such that A 1;j A 1;k 8 k 2 f1: mg. It then assigns job 1 to machine j. Following this, it adds A 1;j to all A i;j 8 i 2 f2: ng. Then, for each remaining job, p 2 f2: mg, it ....

....basis. Although we did not study their performance here, genetic algorithms have been proposed as a good way to schedule tasks on heterogeneous resources, particularly when communication or synchronization is needed between tasks [13] 14] Many systems have followed the lead of SmartNet [6] in implementing intelligent schedulers, such as those we describe here, in their resource management systems [11] 4] 16] 5 Summary In this paper, we experimented with several applications on resources with differing loads and fitted their run times to distributions. We then used these ....

Freund, R., Kidd, T., Hensgen, D., and Moore, L. Smartnet: A Scheduling Framework for Heterogeneous Computing. Proceedings of the International Symposium on Parallel Architectures, Algorithms and Networks (1996).

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