A Performance Comparison Of Four Buffering Schemes For Multistage Interconnection Networks
ABSTRACT Multistage interconnection networks (MINs) are used to connect processors and memories in large-scale multiprocessor systems. MINs have also been proposed as switching fabrics in ATM networks. A MIN consists of several stages of small crossbar switching elements (SEs). A number of buffering schemes are used in the SEs to increase the throughput of MINs and prevent internal loss of packets. the objective of this article is to compare the performance of MINs using different buffering schemes in the presence of uniform and nonuniform traffic patterns. The results obtained from the study will help computer architects and network designers in choosing appropriate buffering strategies for fabric design and configuration of MINs. The normalized throughput, packet loss, and packet mean delay have been used as the performance measures for comparing the different buffering strategies. Results show that the performance of split-shared and output-buffered MINs is considerably better than that of input-buffered MINs when the hot request rate is low. However, the performance is identical for all the buffering schemes when the hot request rate is medium or high.
- SourceAvailable from: G. E. Rizos[Show abstract] [Hide abstract]
ABSTRACT: Omega Networks are a famous subclass of blocking Multistage Interconnection Networks (MINs). They have been recently identified as an efficient interconnectio n network for a switching fabric of communication structures such as gigabit ethernet switch, terabit router , and ATM switch. Interconnection network performance is also a key factor when constructing multiprocessor systems. In this paper we are interested in studying the influence of the blocking mechanisms on the main performance parameters of a typical 8x8 Omega Network with finite buffer size queues. We investigate the packet loss pro bability , the throughput and the latency of an Omega Network using both the Back-pressure and the Block-and-lost blocking Models respectively. This study can be used in future in order to analyse the performance of an actual MIN , where lost packets are resubmitted in the MIN.02/2007;
- [Show abstract] [Hide abstract]
ABSTRACT: In future, multicore processors with hundreds of cores will collaborate on a single chip. Then, more advanced network-on-chip (NoC) topologies will be needed than today's shared busses for dual core processors. Multistage interconnection networks, which are already used in parallel computers, seem to be a promising alternative. In this paper, a new network topology is introduced that particularly applies to multicast traffic in multicore systems and parallel computers. Those multilayer multistage interconnection networks are described by defining the main parameters of such a topology. Performance and costs of the new architecture are determined and compared to other network topologies. Network traffic consisting of constant size packets and of varying size packets is investigated. It is shown that all kinds of multicast traffic particularly benefit from the new topology.Computers & Operations Research 01/2008; · 1.91 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Virtual output queueing is known to overcome the head of line blocking problem of input queueing. This type of buffering is widely used in ATM networks. In order to deliver desired performance, virtual output queueing requires efficient and effective scheduling algorithm with low operating complexity. For large scale switches this might be difficult to achieve, as algorithm complexity increases together with the size of the switch fabric. It is possible to resolve this problem by using interconnection network architectures with distributed buffers. In this case, each network node is a 2×2 switching element that employs virtual output queueing and dedicated selection policy that operates locally. Thus, large scale switches can be achieved without the expense of complex scheduling algorithms. In this paper, performance characteristics of the longest queue first, oldest cell first and random selection policies in the Banyan like interconnection network are studied. Results show that the longest queue first selection policy outperforms others in terms of packet loss performance, whereas random selection policy achieves low throughput-delay ratio performance.