SpaceCurve CEO Slitz has an interesting perspective on public cloud:

SpaceCurve’s platform is designed to run on computer clusters with at least 1,000 nodes, and not in a public cloud because it takes too long to move datasets this large. “It’s like draining Lake Union into the Pacific with a garden hose,” Slitz says.

Comically simple bandwidth bottlenecks continue to be a hassle in big data analytics.