AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
In large parallel and distributed systems, such as networks of workstations or the Internet, the bandwidth of the interconnection network usually is the major bottleneck for the performance of distributed applications

Minimizing Congestion in General Networks

FOCS, pp.43-52, (2002)

Cited by: 249|Views131
EI

Abstract

A principle task in parallel and distributed systems is to reduce the communication load in the interconnection network, as this is usually the major bottleneck for the performance of distributed applications. In this paper we introduce a framework for solving on-line problems that aim to minimize the congestion (i.e. the maximum load of ...More

Code:

Data:

0
Introduction
  • In large parallel and distributed systems, such as networks of workstations or the Internet, the bandwidth of the interconnection network usually is the major bottleneck for the performance of distributed applications.
  • Speaking, the weight function w (X) counts for a subset X, the bandwidth of all edges that are adjacent to nodes in X and that do not connect nodes of the same cluster with respect to the sub-partition corresponding to Vt .
  • Based on this weight function, the authors define for a level node vt of the decomposition tree, the bandwidth-ratio λvt as λvt :=
Highlights
  • In large parallel and distributed systems, such as networks of workstations or the Internet, the bandwidth of the interconnection network usually is the major bottleneck for the performance of distributed applications
  • This is due to the fact that it is often more expensive or more difficult to increase the bandwidth of the interconnection network than to increase processor speed and memory capacities at individual nodes
  • The load has to be distributed evenly among all network resources. This corresponds to minimizing the congestion, i.e., the maximum taken over all network links of the amount of data transmitted by the link divided by the respective bandwidth
  • In this paper we show the somehow surprising result that knowledge of the current load in the network is not needed in order to achieve a polylogarithmic competitive ratio, w.r.t. the congestion
Results
  • Theorem 1 For any request sequence σ for an online problem on network G that can be processed with congestion C the straightforward simulation on TG yields congestion Ct ≤ C.
  • In the first part the authors give a general, i.e., problem-independent, simulation result that relates the expected relative load of any edge of G to the congestion on the tree TG.
  • In the second part the authors show for the data management and the online routing problem that this simulation can be adopted such that the relative load of any edge e in G does not deviate too much from its expectation.
  • The simulation of such a tree edge induces load on e = (x, y) only if both endpoints x and y of e lie in the cluster Sut .
  • This holds because the routing paths between the nodes simulating ut and vt do not leave the cluster Sut .
  • Theorem 6 Given a graph G and an associated decomposition tree TG there exists an oblivious online routing algorithm that is O(log n · h · max{δ, λ})-competitive with respect to the congestion.
  • Remark 7 By utilizing the results of Section 4 on the parameters of TG this theorem gives an O(log3 n)-competitive online routing algorithm for general networks.
  • Theorem 8 There is an O(log3 n)-competitive online algorithm for the data management problem on general networks.
  • A tree TG in which any cluster Svt fulfills both properties has bandwidth-ratio at most λ and weight-ratio at most δ.
  • Lemma 10 Given a level cluster S ⊂ V that fulfills the bandwidth property, it is possible to partition S into subclusters Si with the following characteristics.
Conclusion
  • The cluster V that contains all nodes in the network fulfills the bandwidth-property, because out(V ) = 0.
  • Let for the running time of the algorithm Ru and Rv denote the set in the partition PR that contain the node u and v, respectively.
  • Later the authors will show that if no subset violates this condition the cluster S fulfills the weight-property, i.e., Requirement 4 holds.
Reference
  • W. Aiello, F. T. Leighton, B. M. Maggs, and M. Newman. Fast algorithms for bit-serial routing on a hypercube. Mathematical Systems Theory, 24(4):253–271, 1991.
    Google ScholarLocate open access versionFindings
  • J. Aspens, Y. Azar, A. Fiat, S. A. Plotkin, and O. Waarts. Online load balancing with applications to machine scheduling and virtual circuit routing. In Proceedings of the 25th ACM Symposium on Theory of Computing (STOC), pages 623–631, 1993.
    Google ScholarLocate open access versionFindings
  • Y. Aumann and Y. Rabani. An O(log k) approximate mincut max-flow theorem and approximation algorithm. SIAM Journal on Computing, 27(1):291–301, 1998.
    Google ScholarLocate open access versionFindings
  • B. Awerbuch and Y. Azar. Local optimization of global objectives: Competitive distributed deadlock resolution and resource allocation. In Proceedings of the 35th IEEE Symposium on Foundations of Computer Science (FOCS), pages 240–249, 1994.
    Google ScholarLocate open access versionFindings
  • A. Borodin and J. E. Hopcroft. Routing, merging and sorting on parallel models of computation. Journal of Computer and System Sciences, 30(1):130–145, 1985.
    Google ScholarLocate open access versionFindings
  • T. Hagerup and C. Rub. A guided tour of Chernoff bounds. Information Processing Letters, 33(6):305–308, 1990.
    Google ScholarLocate open access versionFindings
  • C. Kaklamanis, D. Krizanc, and T. Tsantilas. Tight bounds for oblivious routing in the hypercube. In Proceedings of the 2nd ACM Symposium on Parallel Algorithms and Architectures (SPAA), pages 31–36, 1990.
    Google ScholarLocate open access versionFindings
  • P. N. Klein, S. A. Plotkin, and S. B. Rao. Excluded minors, network decomposition, and multicommodity flow. In Proceedings of the 25th ACM Symposium on Theory of Computing (STOC), pages 682–690, 1993.
    Google ScholarLocate open access versionFindings
  • F. T. Leighton and S. B. Rao. An approximate max-flow mincut theorem for uniform multicommodity flow problems with applications to approximation algorithms. In Proceedings of the 29th IEEE Symposium on Foundations of Computer Science (FOCS), pages 422–431, 1988.
    Google ScholarLocate open access versionFindings
  • [11] B. M. Maggs, F. Meyer auf der Heide, B. Vocking, and M. Westermann. Exploiting locality for networks of limited bandwidth. In Proceedings of the 38th IEEE Symposium on Foundations of Computer Science (FOCS), pages 284–293, 1997.
    Google ScholarLocate open access versionFindings
  • [12] L. G. Valiant and G. J. Brebner. Universal schemes for parallel communication. In Proceedings of the 13th ACM Symposium on Theory of Computing (STOC), pages 263–277, 1981.
    Google ScholarLocate open access versionFindings
  • [13] B. Vocking. Static and Dynamic Data Management in Networks. PhD thesis, Universitat Paderborn, Dec. 1998.
    Google ScholarFindings
  • [14] B. Vocking. Almost optimal permutation routing on hypercubes. In Proceedings of the 33rd ACM Symposium on Theory of Computing (STOC), pages 530–539, 2001.
    Google ScholarLocate open access versionFindings
Author
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科