I actually use MariaDB Galera Cluster with 3 master nods. Workload is about 70% read, 30% write and it's very important to have consistent data with every select - this is why I choose Galera 3-masters setup.
But I have problem with disk I/O - each server have 2xSSD drive in Stripping RAID0 and I have about 10% iowait.
What I want - hardware raid10 from all six drives which I have, on separate server and shared for all nodes. Then update/insert can be done only with server which run query and all servers have actual data + six drives raid10 will have many more IOPS.
Is this possible? I read about NDB storage engine but it not work as drop-in and is not possible to rewrite very large application. Next what I found is ScaleDB which looks like exactly what I want but only closed beta is available.