As someone who has designed a few large-scale time-series databases, I read the entire patent. It was filed in December 2012, long after I last did a clean room design, so pretty much everything I've done could be viewed as prior art. I would have to imagine Splunk has something to say about this too.

The primary claim of novelty in this patent is storing index block metadata on a separate server than the storage blocks. To the extent that pretty much every distributed time-series platform ever designed separates index block metadata from storage blocks, it does not reflect positively on the competency of the software team at GE that they think this is novel. (I realize that this may not have been the choice of the software team.)

Furthermore, in properly designed systems, these functions are separated at the granularity of a process rather than a server because it is more flexible and efficient. They can be implemented on separate servers, and usually are, but it is not a requirement because requiring it would be stupid. But if GE did not make this a requirement, I doubt it would have withstood even the nominal prior art search; that limitation on scope was needed to get it through.

Overall, the architecture of the time-series database implied in the patent is consistent with someone that learned their poor distributed system habits from the Hadoop ecosystem and decided to reinvent the time-series database without knowing much about how those systems actually work. However, I am not surprised this made it through the US patent system.