Efficient mechanism for fairness and deadlock-avoidance in high-speed networks

  • Yoram Ofek
  • Moti Yung
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 486)


High-speed network is a new environment, motivated largely by the advance of fiber-optics technology. Such a network cannot operate in conditions which may cause deadlock or may starve certain network nodes since these phenomena result in long waiting times and invocations of delaying recovery procedures which imply (by definition) very low speed.

On the other hand, in order to accelerate the network algorithms, such networks are likely to have the following characteristics: 1. network traffic has higher priority than input traffic (to guarantee fast uniform routing services at intermediate nodes independent of local demand), and 2. packets are removed at destination unlike traditional broadcast approach in loca-area network (to avoid unnecessary communication resource wasting). These two characteristics imply the possibility of overloading a node with network traffic which implies an inherent lack of fairness that cannot be solved by a global mutual exclusion mechanism such as the traditional token (as in a token ring).

In this work we present a general control mechanism for fair access to communication links of a network with an arbitrary topology. In this abstract motivated by high-speed operation, we assume that the network nodes operate in real time, i.e. known links delay. (The work can be translate to other asynchronous models.) The proposed control mechanism operates over a spanning tree that was constructed for this purpose in the network. The fairness mechanism implies deadlock-freeness. Furthermore, it guarantees a stronger condition which is equifairness (equal opportunity is given to every node at each control cycle).

The mechanism regulates the network access globally and can work much faster than previous ones. It is optimal as its time is proportional to the network diameter (previous mechanisms were proportional to the network's size). This is crucial since a shorter control cycle unloads hot spots in a faster rate. Furthermore, the mechanism is self-balancing as it adapts itself to the relative speed of the various parts of the network and it also automatically tolerates one failure per cycle.

As a result, the mechanism can be part of a high-speed network architecture in which, on one hand a node can try to transmit asynchronously, without reservation, as much as it can only by observing the state of its adjacent links, and on the other hand the network access and flow control will ensure no loss, fair access to the network and no deadlocks.

In addition, a clock-driven self-stability property of the algorithm will ensure that local state information and a reliable local clock will be sufficient to recover from any faulty state within a single control cycle.


Span Tree Control Signal Child State Output Link Spatial Reuse 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. [AIP90]
    C. Arbib, G. F. Italiano, A. Panconesi, “Predicting deadlock in store-and-forward networks,” Networks, to appear.Google Scholar
  2. [AKP90]
    B. Awerbuch, S. Kutten and D. Peleg. Efficient Deadlock-free Routing, INFOCOM-91 (to appear).Google Scholar
  3. [ACK90]
    B. Awerbuch, I. Cidon and S. Kutten. Dynamic Tree Maintenance, Found. of Comp. Scie (FOCS) 90., IEEE.Google Scholar
  4. [BT84]
    G. Bracha and S. Toueg. A distributed algorithm for generalized deadlock detection. In Proc. 3rd ACM Symp. on Principles of Distributed Computing, pages 285–301. ACM, August 1984.Google Scholar
  5. [CO89a]
    I. Cidon and Y. Ofek, ”Distributed Fairness Algorithm for Local Area Networks with Concurrent Transmissions,” the 3rd International Workshop on Distributed Algorithms, Nice, September 1989, IBM Research Report RC 15051, October 1989.Google Scholar
  6. [CO89b]
    I. Cidon and Y. Ofek, ”MetaRing — A Full-Duplex Ring with Fairness and Spatial Reuse,” IBM Research Report RC 14961, September 1989, also INFOCOM'90.Google Scholar
  7. [Fr]
    N. Francis, Fairness. Springer Verlag, New York.Google Scholar
  8. [CJS87]
    I. Cidon, J. Jaffe, and M. Sidi. Distributed store-and forward deadlock detection and resolution algorithms. IEEE Trans. on Commun., COM-35:1139–1145, May 1987.Google Scholar
  9. [CM82]
    K.M. Chandi and J. Misra. A distributed algorithm for detecting resource deadlocks in distributed systems. In Proc. 1st ACM Symp. on Principles of Distributed Computing, pages 157–164. ACM, August 1982.Google Scholar
  10. [G81]
    K.D. Günther. Prevention of deadlocks in packet-switched data transport systems. IEEE Trans. on Commun., COM-29:512–524, May 1981.Google Scholar
  11. [GHS83]
    R.G. Gallager, P.A. Humblet, and P.M. Spira. A distributed algorithm for minimum weight spanning trees. ACM Trans. on Programming Lang. and Syst., 5:66–77, 1983.Google Scholar
  12. [Gop84]
    I.S. Gopal. Prevention of store-and-forward deadlock in computer networks. IEEE Trans. on Commun., COM-33:1258–1264, Dec. 1985.Google Scholar
  13. [JS89]
    J.M. Jaffe and M. Sidi. Distributed deadlock resolution in store-and forward networks. Algorithmica, 4, 1989.Google Scholar
  14. [LiRo83]
    M. T. Liu and D. M. Rouse, “A Study of Ring Networks,” Proc. IFIP WG6.4/University of Kent Workshop on Ring Technology Based Local Area Networks, September 1983, pp. 1–39.Google Scholar
  15. [MM79]
    D.A. Menascoe and R. Muntz. Locking and deadlock detection in distributed databases. IEEE Trans. on Software Eng., SE-5:195–202, 1979.Google Scholar
  16. [MS80]
    P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks i: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345–352, March 1980.Google Scholar
  17. [Obe82]
    R. Obermarck. Distributed deadlock detection algorithm. ACM Trans. on Database Syst., 7:187–208, 1982.Google Scholar
  18. [OY90a]
    Y. Ofek and M. Yung, ”Principles for High Speed Network Control: loss-less and deadlock-freeness, self-routing and a single buffer per link,” ACM PODC'90, pp. 161–165.Google Scholar
  19. [OY90b]
    Y. Ofek and M. Yung, ”Lossless Asynchronous Broadcast-and-Feedback on the MetaNet Achitecture,” INFOCOM'91 (to appear).Google Scholar
  20. [Ros86]
    F. E. Ross, ”FDDI — a Tutorial,” IEEE Communication Magazine, Vol. 24, No. 5, May 1986, pp. 10–17.Google Scholar
  21. [SLCG]
    M. Sidi, W. Z. Liu, I. Cidon and I. Gopal, ”Congestion Avoidance through Input Rate Regulation,” GLOBCOM'89, Dallas Texas, 1989.Google Scholar
  22. [TU81]
    S. Toueg and J.D. Ullman. Deadlock-free packet switching networks. SIAM J. on Comput., 10:594–611, 1981.Google Scholar
  23. [Tur86]
    J. Turner, ”New Directions in communications (or Which Way to the Information Age?)”, IEEE Communications Magazine, October 1986, Vol. 24, No. 10.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1991

Authors and Affiliations

  • Yoram Ofek
    • 1
  • Moti Yung
    • 1
  1. 1.IBM T. J. Watson Research CenterYorktown Heights

Personalised recommendations