K. Windisch, V. M. Lo, and B. Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the International Conference on Parallel Processing, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....allocation strategies (e.g. 7, 8, 9, 10] These strategies are dependent on the network topology and the underlying routing algorithm. The work in [11] studied the influence of network contention on application performance due to scatter processor allocation in 2D meshes. However, the work in [12, 13] indicated that contention free allocation is not critical to the application performance. Their study showed that the best performance was exhibited by non contiguous algorithms that provide some degree of contiguity. This is true by considering that software latency is the major contributor to ....

....best system utilization without significantly sacrificing individual application performance. The amount of external contention (or how the allocated processor should be scattered) does not make much difference to the application performance. This result is quite contrary to the result reported in [13]. However, this can be explained by the same argument in adopting gang scheduling in multiprocessors. ffl In systems with low software latency, contention free allocation policies should be considered. ffl In order to accommodate applications with varying communication characteristics, we ....

K. Windisch, V. Lo, and B. Bose, "Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes," Tech. Rep. Technical Report, University of Oregon, 1995.

....non contiguous, based on whether or not the set of allocated processors are directly connected by links in the interconnection network. Research in job scheduling and processor allocation is thoroughly surveyed in [8] more recent work in this area includes that reported in [10] as well as our own [17, 22, 18]. This project is distinguished from our previous work in that we evaluate the experimental method, not the scheduling techniques themselves. Some of the first researchers to use real traces from production machines to drive their simulations include [19, 1, 18] At the same time, analysis of ....

K. Windisch, V. M. Lo, and B. Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the International Conference on Parallel Processing, 1995.

....only 34 to 66 are reported [9, 18, 8, 10] In contrast, the class of noncontiguous allocation strategies allocates nodes that are dispersed throughout the system. Fig. 3 shows examples. They experience no fragmentation and thus outperform contiguous strategies reaching utilizations of up to 78 [10, 15, 14]. To further improve the performance of noncontiguous strategies, it is necessary to select allocations that cause minimal message passing contention. External contention is due to job interference: communication between dispersed nodes of a given job may require a communication link that is ....

....problems. This is unfavorable and contradicts the goal of high throughput over a stream of jobs. 2. 2 Non contiguous allocation strategies and contention To utilize unallocated nodes that are not necessarily contiguous, non contiguous processor allocation strategies have been proposed [7, 10, 15, 14]. They are also known as scattered or fragmentation free allocation strategies. Nodes allocated to the same job are allowed to be dispersed throughout the interconnection network. Fig. 3 shows examples. What are the effects of dispersed nodes On one hand, there are several advantages: Neither ....

[Article contains additional citation context not shown here]

K. Windisch, V. Lo, and B. Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the International Conference on Parallel Processing, 1995.

....non contiguous, based on whether or not the set of allocated processors are directly connected by links in the interconnection network. Research in job scheduling and processor allocation is thoroughly surveyed in [8] more recent work in this area includes that reported in [10] as well as our own [16, 21, 17]. This project is distinguished from our previous work in that we evaluate the experimental method, not the scheduling techniques themselves. Some of the first researchers to use real traces from production machines to drive their simulations include [18, 1, 17] At the same time, analysis of this ....

K. Windisch, V. M. Lo, and B. Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the International Conference on Parallel Processing, 1995.

....capability is the pre x based Buddy scheme. We note that Cyclic with contiguous masks supports n times the recognition capability of pre x based schemes. Also shown is the recognition capability of three schemes not considered in this paper: Gray Code and Double Gray Code [2] and Partners [1, 18]. These schemes use non contiguous masks but do not allow for all possible patterns of non contiguity. Interestingly, they all do worse than Cyclic. This is why we chose Cyclic for recognition in the CBA address allocation scheme. We originally designed Cyclic as a subcube allocation algorithm ....

Kurt Windisch, Virginia Lo, and Bella Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the 1995 International Conference on Parallel Processing, pages II{164{II{168, 1995.

....three non contiguous processor allocation strategies: Random, Paging, and the Multiple Buddy Strategy (MBS) with three well known contiguous allocation schemes: Frame Sliding, First Fit, and Best Fit. These strategies represent a continuum with respect to degree of contiguity. As we have shown in [29], these strategies are also directly applicable to processor allocation in k ary n cubes which include the hypercube and torus. One of these strategies, the Multiple Buddy Strategy (MBS) is an elegant algorithm we have developed that exploits the advantages of non contiguity with respect to ....

K. Windisch, V. M. Lo, and B. Bose. Contiguous and non-contiguous processor allocation algorithms for k-ary n-cubes. In Proceedings of the International Conference on Parallel Processing, 1995. To appear.

....contention due to non contiguous allocation is not as serious as the fragmentation effects of the contiguous allocation. Thus, we observed that non contiguous allocation performed better overall, even when message passing contention was considered [8] We have run similar experiments, described in [10], to determine the feasibility of non contiguity in k ary n cubes with respect to network contention. 7 Conclusion We have presented several new contiguous and noncontiguous processor allocation algorithms, for k ary n cube architectures. Due to the great versatility and generality of the k ary ....

....It was also interesting to note that contiguous schemes performed better on the high dimensional 1024 node k ary n cubes we tested, such as binary hypercubes, than on low dimensional ones, such as 2D tori. Performance with respect to message passing contention is further discussed and analyzed in [10]. Our ongoing work in this area includes examining the performance of a variety of job scheduling policies when used in conjunction with both contiguousand non contiguous allocation schemes. Our goal is to develop efficient and practical resource management systems that fully utilize the ....

K. Windisch, V. Lo, and B. Bose. Contiguous and noncontiguous processor allocation algorithms for k-ary n- cubes. Technical report, University of Oregon, 1995.

....external fragmentation. Utilizations of only 34 to 66 are reported [2, 10, 1, 3] The class of non contiguous allocation strategies allocates nodes that are dispersed throughout the system. They experience no fragmentation and thus outperform contiguous strategies reaching utilizations of 78 [3, 8, 7]. To further improve the performance of noncontiguous strategies, it is necessary to select allocations that cause minimal message passing contention. External contention is due to job interference: communication between dispersed nodes of a given job may require a communication link that is ....

K. Windisch, V. Lo, and B. Bose. Contiguous and noncontiguous processor allocation algorithms for k-ary n- cubes. In Proceedings of the International Conference on Parallel Processing, 1995.

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