Title:Performance
Analysis of Multistage Interconnection Networks determining optimal parameters
for data-intensive business applications
Abstract: Multistage Interconnection Networks (MINs) are
frequently used for connecting processors in parallel computing systems or
constructing high speed networks as ATM (based on Asynchronous Transfer Mode)
and Gigabit Ethernet Switches. Nowadays there is a great interest about Switching
Systems and especially for self-routing systems called Banyan Switches. New
business applications require distributed computing implementations , but
old networks are too slow to allow e±cient use of remote resources.
In addition to , multimedia are considered as applications with high bandwidth
requirements. Some of them are also sensitive to packet loss and claim reliable
data transmission. Special applications require bulk data transfers for database
replication or load balancing and therefore packet loss minimisation is necessary
in order to increase the performance of them. The demand for high performance
multimedia services such as full motion video on demand is becoming an increasingly
important driving force in the communication market in the Digital Age. So
, the performance of MINs is a crucial factor , which we have to take into
account in the design of new applications. The performance of them is mainly
determined by their communication throughput and cell latency, which have
to be investigated either by time-consuming simulations or approximated by
mathematical models. In this paper we investigate the performance of Banyan
Switches, one of the most famous classes of MINs , in order to determine
optimal values for hardware parameters under deferent conditions. We study
a typical 8x8 Banyan Switch with internal queuing under variable length buffer
size using the Back-pressure blocking Model which is more realistic approach
than using the Block-and-lost Model. This study can also be used in future
in order to analyze and evaluate the use of MINs as an intercommunication
medium especially in symmetric multiprocessor systems , which will able to
satisfy future data-intensive business applications.
Authors: D.C.
Vasiliadis, G.E. Rizos, and C. Vassilakis