Comment by justinsb

Can someone explain why the in-memory caching is such a big win? Does Hadoop MapReduce not do caching as well? I'd expect at least filesystem caching when the computation is running on the same machine as the data block...