Nice. Of course the approach quickly hits a limitation if the dataset is too big to be replicated on every app server, or if low-latency updates are required. But there is wisdom in knowing what is and what isn't important in your particular application, and stripping away everything that isn't. Once you've taken away everything that isn't needed, the resulting architecture may end up being startlingly simple.
Of course the approach quickly hits a limitation if the dataset is too big to be replicated on every app server, or if low-latency updates are required.
This is handled by grouping nodes together physically, and calling the resulting bundle a datacenter.
Data is stored using a Cassandra-like algorithm, that locates the data on each node in the cluster.
This solves both the "too much data for one machine" and the low-latency problem. Now the data just needs to fit in a single datacenter.
