A Comparison of Cache Performance in Server-Based and Symmetric Database Architectures

We study the cache performance in a symmetric distributed main-memory database. The high performance networks in many large distributed systems enable a machine to reach the main memory of other nodes more quickly than the time to access local disks. We therefore introduce remote memory as an additional layer in the memory hierarchy between local memory and disks. In order to appreciate the tradeoffs of memory and cpu in the symmetric architecture, we compare system performance in alternative architectures. Simulations show that, by exploiting remote memory (in each node’s cache), performance improves over a wide range of cache sizes as compared to a distributed client/server architecture. We also compare the symmetric model to a centralized-server model and parameterize the performance tradeoffs.