Interesting. It's good to see that the stuff some of us were trying to do quarter of a century ago has finally been done.
Oracle's OPS was an attempt to do this back in the 80s and 90s. Also ICL worked on a massively parallel system in the late 80s, and built a relational engine on it which got quite good benchmark results, but senior management decided it didn't want a home-grown RDBMS so that was killed off. Then we were asked to put Ingres on a massively paralllel platform but I killed that idea fairly quickly - we had an Ingres port to our mainframes, and none of the engineers thought that Ingres would be a good starting point for serious parallelism; in the early 90s ICL and Bull and some others were collaborating on another new RDBMS to run on the European Declarative System project's engine - massive parallelism using cheap hardware again - but that went nowhere, as might be expected with a "collaboration" where the partners all had different objectives, so eventually ICL and Oracle collaborated to put Oracle's OPS on ICL's parallel platform (ICL providing the OS and the lock manager, Oracle providing the RDBMS apart from the lock manager) - and senior management decided as soon as it sort of worked to put the project into QA and split up the development team into other projects so that it could never more than sort of work and they could scrap it - despite which ICL France hijacked a pre-production demo system, got it working better than just sort of, and sold it (in 1995) to a big French bank (who were very happy with it) - that success of course didn't prevent the project being scrapped and no more systems were built or sold.