Here is a wiki link to N-body simulation: http://en.wikipedia.org/wiki/N-body_simulation.

This may seem a little off topic, but we know that even a small perturbation may introduce a huge difference to the system and the step granularity is limited in the simulation, so how could we verify the simulation result anyway?

Question: What are some other memory-constrained scaling examples?

Google's search cache maybe? I guess that is similar to Memcached.

matrix multiplication? since we can always get more processors to run if the matrix gets bigger.

@HingOn Matrix multiplication is an O(n^3) algorithm though. You will probably run out time way before space.

Biosimulation maybe?

"Apache Spark: Fast Queries on In-Memory Datasets"

Spark is like Hadoop, but designed for pretty-big-data (small enough to fit in memory on some large number of nodes, versus hadoop's really-big data stored on slow-and-cheap dist).

Scale the number of nodes, scale the size of your data set.

The idea was sparked (teehee) by a Google paper/tool whose name escapes me. Knife? Dagger? Something pointy?

Redis is an in-memory database which usually serves as cache in web, just like what memcached does.