From nobody@cse.Buffalo.EDU Sat Jan 30 15:30 EST 1999 From: Nobody Date: Sat, 30 Jan 1999 15:30:07 -0500 (EST) To: techreps@cse.Buffalo.EDU Subject: techrep: POST request Content-Type: text Content-Length: 2113 ContactPerson: lee-d@cse.buffalo.edu Remote host: dnoces.cse.buffalo.edu Remote ident: lee-d ### Begin Citation ### Do not delete this line ### %R 99-01 %U /projects1/lee-d/dissertation/Dissertation.ps %A Lee, Chain-Wu %T TERRESA: A Task-Based Message-Driven Parallel Semantic Network System %D January 30, 1999 %I Department of Computer Science and Engineering, SUNY Buffalo %K parallel computation, artificial intelligence, semantic network %X The ultimate goal of research in artificial intelligence is to build a system with human intelligence. Yet, despite considerable effort, artificial intelligence systems are still a long way from that goal. Various reasons attribute to why the goal cannot be achieved at current stage, one of the reasons is the vast amount of information to be processed. When presented with real world domains of millions of concepts and assertions, almost all existing systems are unable to deliver the results within a reasonable amount of time. This dissertation describes a parallel semantic network system, TERRESA, based on SNePS. TERRESA is a task-based, message-driven, multithreaded parallel semantic network system with the support of shared knowledge, load balancing and duplicate checking on MIMD distributed memory/message passing systems. TERRESA can be split into two collaborating components: the host module and the slave module. The host module interacts with the user and processes the information for the slaves, while the slave module performs task execution. The current version of TERRESA focuses on SNePS path-based inference only. TERRESA is designed to be portable among different parallel machines, and therefore most of the system dependent functions are either avoided or re-implemented. TERRESA is written in ANSI C and MPI (Message Passing Interface) library with the help of flex lexical analyzer and yacc parser generator. Porting TERRESA to a new platform only requires recompiling in most cases. The performance studies of TERRESA have been done on SUN multiprocessor server.