re: State Street Private Cloud: $600 Million Savings Goal
Everybody thinks that "cloud" (i.e. Hadoop or some other form of global file system with balanced IO) is the wave of the future. I remain unconvinced, but interested. The problems are that writing applications for a global cluster is really complex and, if you don't do it perfectly, you can create some real latencies. The other issue I have is that when you are working with a series of different, often open source, technologies in the OS stack, there is very little coordination of patch/upgrade release schedules or testing for your specific config. I have seen several clusters get taken down because someone patched a hypervisor, for instance, and it had some unknown inconsistency with firmware that was running on a switch that knocked a good portion of the cluster off-line. The complexity goes way up with this model. You are essentially building your own supercomputing operating system with some help from open source. I don't see anyone reducing on-going operational costs by standardizing on this model vs. some flavor of Unix. Hardware costs, yes, but I don't see how there will be less to manage with this model.