ooh, ok. you might also check out openshmem (http://openshmem.org/site/) and openucx (http://www.openucx.org). I'd been planning on integrating both in RaftLib just not enough cycles to get it done yet. These combined would make it much easier to maintain a relatively portable yet performant back end. My thesis research was all about locality, memory placement, and throughput for big data systems. Current work is similar but I'm not neck deep in the hardware dev world. There are current research efforts on my part outside of work, most center around the raftlib.io platform. Before I forget, you might also want to check out the graph partitioning frameworks like metis and scotch...both are used in some MPI frameworks for more optimal partitioning. To get topology data you might want to look at the hwloc framework, it's cross platform and provides input for things like NUMA/cache/PCIe topology for optimization. I haven't had a chance to integrate this hook into RaftLib, however it's just a few lines away once I find the time. If you're wondering...I started out writing my own fiber library for RaftLib. Had ports for both IBM Power and Intel, but it gets a bit tiring maintaining/optimizing for every new architecture. Qthreads and the like have been used in HPC circles for quite awhile, so it made sense. There was no way I was beating them for dev time, so might as well join them.

Based on your auto-tuning discussion...RaftLib aims to do something similar, but for big-data stream processing workloads. Here's my 2016 IJHPCA paper: http://hpc.sagepub.com/content/early/2016/10/18/109434201667...

It looks like it's behind a paywall so if you don't have access I'll update my website with the "author archive" copy sometime today...will be at ( http://jonathanbeard.io/media ) once I update it. Bottom line if there's intersected interest, definitely open to collaboration :).