DeWitt and Gray compare shared-memory (shared-everything), s…
DeWitt and Gray compare shared-memory (shared-everything), shared-disk, and shared-nothing architectures for parallel database systems. Assume you build a analytical DBMS for large, partitionable scan-join workloads with minimal skew, and each node has its own CPU, memory, and disks connected via a network with reasonable bandwidth. Under these assumptions, which architecture is expected to deliver the best scalability (near-linear speedup) as you add more nodes?