Information processing with associative processors

Computer architectures for logic-oriented data/knowledge bases

“…Prolog machines are designed to optimize the performance in executing compiled logic programs by exploiting fast registers and pipelining techniques. In this section, we consider associative processors HAS (Hahne et al, 1985) and ASSIP-T (Schneider and Dilger, 1986), and high performance Prolog machines PLM (Dorby et al, 1984(Dorby et al, , 1985 and IPP (Abe et al, 1987).…”

“…Figure 9 ASSIP-T (Schneider and Dilger, 1986) corresponding to <1, 1>, and thus 6^ is considered as a result. The term q(X, Y) is found for <2, 1>, and the substitution 6i is applied to the term producing the subgoal <-q(a,Z).…”

“…Another computer, ASSIP-T, consists of a master processor, two associative memories (CAM1 and CAM2), and multiple inference and unification processors (Schneider and Dilger, 1986) ( Figure 9). The main principle of ASSIP-T is to separate unification operations from resolution by maintaining two graphs called a deduction plan and a unification graph with constraints (UwC).…”

“…Parallelism certainly plays the most important role in designing these machines. For data intensive operations, there exists opportunities for performance improvements that can be achieved (Robinson, 1985;Stormon, 1986;Woo, 1985a) -Stream Unification Machines (Shobatake and Aiso, 1986;Tanabe and Aiso, 1987) Resolution Machines -Associative Processors (Hahne et al, 1985;Schneider and Dilger, 1986) -Prolog Machines (Abe et a., 1987;Dorby et al, 1984Dorby et al, , 1985Fagin and Despain, 1987) • Database Machines Deductive Database Machines (Li, 1984;Sakai et al, 1986;Tanaka, 1986) Database Machines for Unnormalized Relations (Lipovski, 1978;Su et al, 1979) • (Shaw, 1985;Stolfo, 1987;Thinking Machines Technical Report, 1987) Intelligent Memory machines (Wah et al, 1984;Yokota and Itoh, 1986) Term/Clause Indexing Contention Free Shared Buffer by exploiting parallelism since the data rate from slow secondary storage can be increased proportionally to the number of disk systems working in parallel. However, for inferencing, parallel processing does not gain much speed up.…”