More importantly Red Hat has deprecated FCoE in RHEL, which is big news, because at a previous $JOB they went all in on FCoE because it was supposed to be the future.
Expensive hardware, with little gain sadly. It was a nice idea, however at its very core is a fairly large problem: converged network adaptors are problematic.
Unless you have lots of bandwidth in said adaptor (ie 56gig inifiband) you are going to get contention between network and disk IO.
I agree, not only that, but at least with Cisco gear FCoE was a real pain in the behind to manage and configure. So much duplication in configuration and settings across a lot of gear, and it never was as smooth as they made it out to be.
We've been big Cisco customers for years and ever since I saw FCoE I thought it was a disaster. All of the DCB extensions that had to go into ethernet to get it to work was such an ugly mess. It was just too complicated compared to alternatives, and iSCSI got a free ride on all that work (ethernet pause, flow control, etc) and was far simpler to implement. And of course with lots of 10Gb options with iSCSI offload, it was getting harder and harder to find any advantage LARGE ENOUGH in FCoE to justify it's cost and complexity.
Completely agreed, but it is something that Cisco was still pushing fairly heavily alongside their partners. Various different vendors bought into it, and it was deployed heavily in various different MSP's.
The complexity of FCoE is staggering, and the configuration required across all the different moving pieces to make it a success made things even more difficult!
NVME over fabrics.. I guess, the general idea being to run NVME over FC, RCoE, or infiniband.
Given the existing install-base of FC, I'm guessing as people upgrade to the 32/128Gbit adapters they will start to purchase disk's that can support FC-NVME as well. Which will bootstrap the market there.
Although it could go to infiniband as well, if people buy into the converged infiniband/ethernet adapter route.
To soon to tell, but a lot of it will be dependent on which technology does a better job avoiding the "forklift" upgrade problem that FCoE required.
Hyperconverged Infrastructure is definitely eating a lot of the traditional storage vendors lunches. Using Ubuntu with JuJu/Ceph/OpenStack all on the same servers provides plenty of power while reducing costs.
Even VMWare has come on board with vSAN which pushes out vendors like EMC/NetApp because you no longer need them when you can just create it against your existing hypervisors. Sure you can run one or two less VM's on it, but you have less cost overall.
iSCSI off-load that is available on a variety of network cards, along with network policies that are similar to FCoE (never drop a packet) allow you to get the same speed/reliability as FCoE for a lot less.
FCoE requires that the networking gear drops only one packet in 10 million or something like it, if you can make the same guarantees for iSCSI, it is for all intents and purposes the same thing. With iSCSI off-load it is even better.
iSCSI also runs across your existing network stack, and doesn't require purchasing special equipment and is better supporter across a variety of different vendors, thereby making it easier to find the gear with the features you need, rather than settling for something that supports FCoE.
