Implementation and analysis of a packet telephone system
T.O. Brunner · 1989
The results of two studies concerning voice services on an integrated packet switched network are presented. The design of a packet telephone system, for operation on an integrated network is studied through implementation of a prototype. The buffering requirements of a proposed generic Packet Voice Messaging System (PVMS) are determined through an analytical study. The packet telephone system was designed with the following unique features. (1) It was implemented on an Integrated Local Area Network (ILAN), MAGNET I, which supports voice communications by providing a service class tailored to the transmission requirements of voice. (2) The packet telephones were designed to be able to support bi-directional transmission of information coded at a higher bit rate than voice standard. (3) They are designed for integration into a SUN workstation, to enable the implementation of additional voice services. The packet telephone system was structured to facilitate experimentation. Therefore, the design was kept flexible, and many functions were implemented in software. In implementing telecommunications software, the dual requirements of efficiency and modularity dictated a functional separation that was later found to follow an Integrated Reference Model (IRM) for communications protocols. The functions specific to packet voice (packetization, reconstruction, and Speech Activity Detection (SAD)) were streamlined, and implemented with some novel schemes. An architecture for a generic PVMS is proposed, consisting of a disk based storage system, and a separate queue for the packets of each speaker storing a message. A key issue for the designer of a PVMS is buffer sizing, and important system parameters include the number of system users, and the size of the disk block. A study is presented of the buffer occupancy distribution, in order to evaluate buffer overflow, and the resulting degradation of the stored voice quality. Two service disciplines are selected, the Strict Round Robin (SRR) and the Modified Round Robin (MRR). A simulation model is developed, and a simulation study conducted. In order to gain better insight into buffering problems, several analytical models of buffer occupancy were developed. They use some novel modelling techniques to accurately predict buffer occupancy distributions, over ranges of parameters of interest to the system designer.