I'm trying to figure out how to build
Linbit's DRBD 9.0.1 with RDMA support under CentOS 6.7 to see if it's possible to get near NVMe replication performance between two Intel P3700 800GB drives across the two QSFP+ peer to peer connection in two different servers. Specifically to address latency issues that two 10Gb ethernet adapters can't handle due to the TCP stack. My hope is (
based on reading Linbit's performance testing with RDMA) is that the backend storage was their limiting factor.
I don't know if this kind of stuff interests anyone here for their day to day work. From my experiences so far with DRBD it's a pretty neat piece of software for replicating block devices between systems. I can write up the steps to get this working in another thread if it interests anyone?