%T Transputer performance issues using the trollius operating system
%A  James R. Jr. Beers,  Ros Leibensperger,  Moshe Braner,  David Fielding
%E  Hussein S. M. Zedan
%B OUG\-13: Real\-Time Systems with Transputers
%X The Trollius Opeiating System for distributed multicomputer
   offers different types of message\-passing services,
   incorporating different amounts of overhead. Each of the
   operating system activities required for interprocess
   communication is designed as an individual module or layer.
   The different types of Trollius communication can be
   accessed by turning operating system services on or off, or
   by directly utilizing different layers of
   communication.Basically, there are three layers of internode
   communication. The network layer can be used from any node
   to any other node; the data link layer is for
   nearest\-neighbor communication; and Ihe physical layer uses
   the hardware in the most efficient way possible, but
   requires exclusive use of the link. A separate level, the
   kernel level, is used for intranode communication. Operating
   system services that can be turned on or off include
   buffering and virtual circuits.Two different types of
   benchmarks are used to evaluate the performance of the
   different types of Trollius message\-passing. The first type
   calculates message\-passing lime as the sum of the limes
   required for each individual service or component of Ihe
   process. The decrease in message\-passing lime obtained when
   turning off services can ihus be easily measured. The second
   benchmark measures throughput, the amount of information
   that can be sent from one node to another in a given amount
   of time, for different layers of message\-passing, utilizing
   different sendees. This is the more accurate measure when
   series of messages are sent from one node to another. The
   throughput measurements demonstrate the value of the
   physical layer and of virtual circuits.