Performance of a Parallel Transport Subsystem Implementation

Introduction The concept of parallelism may significantly increase the performance of transport subsystems [1, 2, 3, etc.]. A transport subsystem covers a subset of a complete transport system, which combines the functions of the four lower layers according to the OSI Reference Model. Because parallelising standard protocols like OSI TP4/CLNP or TCP/IP suffers from their inherent sequential structure, the transport subsystem PATROCLOS (parallel transport subsystem for cell based high speed networks) realises a parallel protocol architecture with fine granularity adequate for parallel implementations on hybrid multiprocessor platforms. PATROCLOS overcomes the performance limitations by the integration of protocol and implementation issues and by providing the required functionalities between the media access layer interface of cell-based networks and the transport service interface (functionality of the OSI layers 2b-4) in a single protocol component. The approach followed in the PATROCLOS design combines a very fine granularity with a protocol function oriented decomposition into basic modular building blocks. In addition, several characteristics of emerging cell based networks such as ATM and DQDB influence the PATROCLOS protocol design [4]. PATROCLOS provides a number of protocol mechanisms to support different application and network requirements. Especially, mechanisms appropriate for high-speed network environments with large a bandwidth-delay-product have been integrated. PATROCLOS combines a modular design and the avoidance of layering to achieve a high degree of parallelism. Redundant placement of protocol functions in several layers can be avoided, e.g., segmentation and reassembly functions in layer 3 and 4 of the OSI Reference Model. Furthermore, the modular protocol function oriented design simplifies the configuration of the protocol. Configuration facilities will play an important role in future protocols to allow adaptation of protocols to a wide variety of application requirements and different network environments [5, 6]. Moreover, the modular protocol design supports an easy mapping on multiproces-sor architectures, if the protocol components can be used as processes that can run on separate processors. Based on the analysis of XTP [7] specification, version 3.4 [8], which is based on a set of parallel finite state machines (FSMs), and the performance results of the corresponding multi-processor implementation, several design guidelines for protocol archi-tectures appropriate for parallel processing have been derived [9]: reduced data dependences, decoupled control and user data processing, modular function-oriented decomposition, reducing event signalling, minimising common resources. These design guidelines resulted in the PATROCLOS specification. Parallel FSMs are the building blocks of the protocol architecture. Together, all FSMs form a …