Feature request: daemon overprovisioning
In order to improve the scores we are getting in the IO500 benchmark, it would be very interesting to be able to overprovide the daemons in a job run. It has come to our attention that in the NEXTGenIO prototype we will (soon) have dual NVRAM mount points on separate buses (1.5TiB each), as well as dual network cards (and interfaces) on each compute node. This opens up the possibility to run two independent GekkoFS instances at each compute node to run the benchmark and increase performance.
There is a lot of interest right now in being able to improve the IO500 results, @marcvef and @toto, what changes would be needed to be able to support this? To make it simples, assume that the namespaces/dataspaces don't need to be coordinated since we only need to support running the benchmark (cc @rnou).