Cache memory is small in size, fast buffer memory that is used in computer systems for the temporary storage of data of the main memory. Given that disk access is 40-200 thousand times slower than memory access, it is obvious that cache mamagement must aim at diminishing read/write operations from/to the disk, so that the system's response time is restrained. Despite the fact that the operating systems of modern computer systems have complete cache management mechanisms, database management systems have their own mechanisms in order to take advantage of their special data access characteristics. Metrics methods are used to study the performance of cache management mechanisms in database management systems. These metrics can involve simulator models, benchmarks or experimentation and metrics' aquisistion on the real system. The Semantic Index System, that was developed in the Institute of Computer Science, FORTH, manages knowledge bases in which knowledge is modeled in the form of a semantic network. The Semantic Index System deployes its own cache management mechanism. The study of the cache management mechanism's behaviour in the Semantic Index System is the goal of this work. The method of experimenting and collecting metris on the real system was adopted for this study. The experiments on the Sematic Index System were conducted for three different databases, chosen based on their special characteristics, and typical system operation scenarios were used. The parameters of the experiments have been the sizes of the separate mechanisms that cache memory consists of. Total memory consumption and hit rates have been the outcome of the experiments. Interesting conclusions have been drawn concerning the cache management mechanism, as a result of the abovementioned work, as well as performance curves of the separate mechanisms that make up the cache memory given their sizes, during the system's typical usage. These curves can be used for determining the size parameters of the mechanism with the aim of restraining the system's total memory consumption, without allowing its performance deteriorate.