Main Memory Databases vs. Disk-Resident Databases

Storing all database data in memory is an idea that many researchers have been studying it from mid-1980s when RAM prices decreased while their capacity increased. Main Memory Database systems (MMDB) are an efficient solution to store all database data in main physical memory. Conventional database systems like relational databases are optimized for disk I/O operations but Memory resident databases use different optimization mechanism to organize and cluster data in memory in an efficient manner that minimizes disk I/O operations. In this empirical study we compared performance of main memory databases with conventional database systems. For this purpose we chose SQL Server and PERST which are relational disk based and objectoriented main memory databases respectively. Then we evaluated them based of TPC-B and included our experimental results in this paper to assess their behavior from different aspects. Our results show that main memory databases can handle huge amount of data more efficiently than disk-based conventional databases because they have less over head on the system that they are running transaction on.