Nonuniform Traffic Spots (NUTS) in Multistage Interconnection Networks

Multistage interconnection networks (MIN's) form an important class since they provide a compromise in cost/performance between the shared bus and crossbar. Previous performance results have indicated that buffered MIN's provide good performance (low delay) at moderate traffic loads. Those results, however, assumed uniform traffic, in which the destination address is a random variable with uniform distribution. This assumption was attractive since it makes analysis mathematically tractable. We use a detailed simulation model to show that the performance of blocking and discarding MIN's is degraded when the input traffic is not uniformly distributed, producing nonuniform traffic spots (NUTS) in the network. NUTS can occur for various reasons, including an application-specific reference pattern, a small real-time traffic component, or a fault in the network. Degradation occurs because of an imbalance in the load carried by particular internal switches. With blocking control, a tree of saturated queues rooted at the NUTS forms causing degradation for a subset of sources. Discarding does not help since resent packets follow the same congested path and are therefore likely to be discarded again. Other previously proposed solutions to similar problems are also shown to be ineffective in this case. We propose the use of a new control strategy, called diverting, in which buffer contention is handled by routing overflow packets to a (wrong) queue with available space. The diverted packets eventually arrive at an incorrect destination, where they are resubmitted into the network. Since the source node is different in the second pass, the packet can avoid the NUTS where the original diversion took place. Moreover, switches with additional links, originally proposed for fault tolerance, are shown to be very effective in reducing the degradation caused by NUTS. We combine alternate path switches with diverting control for best results.

[1]  Janak H. Patel Performance of Processor-Memory Interconnections for Multiprocessors , 1981, IEEE Transactions on Computers.

[2]  Gregory F. Pfister,et al.  “Hot spot” contention and combining in multistage interconnection networks , 1985, IEEE Transactions on Computers.

[3]  Howard Jay Siegel,et al.  The Extra Stage Cube: A Fault-Tolerant Interconnection Network for Supersystems , 1982, IEEE Transactions on Computers.

[4]  V. P. Kumar,et al.  Design and analysis of fault-tolerant multistage interconnection networks with low link complexity , 1985, ISCA 1985.

[5]  David A. Padua,et al.  Interconnection Networks Using Shuffles , 1981, Computer.

[6]  Kevin P. McAuliffe,et al.  RP3 Processor-Memory Element , 1985, ICPP.

[7]  Debasis Mitra,et al.  Randomized Parallel Communications , 1986, ICPP.

[8]  Howard Jay Siegel,et al.  Interconnection networks for large-scale parallel processing: theory and case studies (2nd ed.) , 1985 .

[9]  Kai Hwang,et al.  Computer architecture and parallel processing , 1984, McGraw-Hill Series in computer organization and architecture.

[10]  Daniel Gajski,et al.  CEDAR: a large scale multiprocessor , 1983, CARN.

[11]  Duncan H. Lawrie,et al.  Access and Alignment of Data in an Array Processor , 1975, IEEE Transactions on Computers.

[12]  Kevin P. McAuliffe,et al.  The IBM Research Parallel Processor Prototype (RP3): Introduction and Architecture , 1985, ICPP.

[13]  Yuval Tamir,et al.  Support for high-priority traffic in VLSI communication switches , 1988, Proceedings. Real-Time Systems Symposium.

[14]  Rosanna Lee On “hot spot” contention , 1985, CARN.

[15]  Marc Snir,et al.  The Performance of Multistage Interconnection Networks for Multiprocessors , 1983, IEEE Transactions on Computers.

[16]  Manoj Kumar,et al.  The Onset of Hot-Spot Contention , 1986, ICPP.

[17]  Nian-Feng Tzeng Design of a Novel Combining Structure for Shared-Memory Multiprocessors , 1989, ICPP.

[18]  Nian-Feng Tzeng,et al.  A fault-tolerant scheme for multistage interconnection networks , 1985, ISCA '85.

[19]  Robert H. Thomas,et al.  Behavior of the Butterfly Parallel Processor in the Presence of Memory Hot Spots , 1986, ICPP.

[20]  Roy M. Jenevein,et al.  Traffic analysis of rectangular SW-banyan networks , 1988, [1988] The 15th Annual International Symposium on Computer Architecture. Conference Proceedings.

[21]  Cauligi S. Raghavendra,et al.  Fault-Tolerant Routing in Multistage Interconnection Networks , 1989, IEEE Trans. Computers.

[22]  Daniel M. Dias,et al.  Packet Switching Interconnection Networks for Modular Systems , 1981, Computer.

[23]  Ralph Grishman,et al.  The NYU Ultracomputer—Designing an MIMD Shared Memory Parallel Computer , 1983, IEEE Transactions on Computers.

[24]  E. W. Mayr,et al.  Simulation of an ultracomputer with several hot spots. Technical report , 1986 .

[25]  Leslie G. Valiant,et al.  A Scheme for Fast Parallel Communication , 1982, SIAM J. Comput..