Storagebod has posted a great reply to Hu Yoshida. You should definitely read both posts; because this is an excellent example of the kind of confusion you can get by reusing existing standard names. In this case, the problem is the “E” in DCE and FCoE. Everybody knows Ethernet technology; it’s used all over the world from private networks people run at home to the largest data centers.
As such, most people have heard their share of horror stories about broadcast storms, packet loss, and other problems associated with these networks. And most people know how to plug in a network cable, which to some equates to “building a network”. So by using the Ethernet label for storage devices and protocols either of two things can happen:
- Many people will think that storage just got easier. Fibre Channel sounds complicated, and having to support a separate networking technology for your storage system doubly so. Since any IT pro knows (or rather, thinks he knows) how to build and maintain an Ethernet network, this leads to, well, interesting setups.
- On the other hand, storage professionals think a protocol/technology that works reliably (FC) will now be used on unreliable networks, wreaking havoc on their carefully designed systems.
In reality, neither of these are true. Fibre Channel stays the same, even when used over ethernet. The only difference between traditional FC and FCoE devices is that you might be able to get away with using a single network for both your storage systems and other networking needs. If used on a typical existing Ethernet network, you might have issues like congestion and packet loss. Which is nothing to be worried about; iSCSI works just fine over networks with packet loss. Performance might suffer, but if your network is already over saturated storage performance is the least of your worries.
FCoE is less forgiving when packet loss is seen, but 10 Gbps networking is required by all vendors that I know of; this reduces the chances of having congestions in today’s environments, and for any serious deployment you’ll be using a DCE network instead of “traditional” ethernet.
The main advantage of Data Center Ethernet is that it includes the IEEE 802.1Qbb Priority-based Flow Control mechanism, designed to “ensure zero loss under congestion in DCB networks”. There’s also IEEE 802.1Qau enhanced congestion notification, which should further reduce the chance of having performance issues on networks under high load.
So in nearly all environments, FCoE should be a viable alternative for Fibre Channel networks. Whether or not it will provide a good Return on Investment at this moment is still up fur debate; if you already have a Fibre Channel infrastructure, replacing it with DCE-compatible equipment and upgrading your storage systems to support FCoE will probably be more expensive than running separate data and storage networks for the foreseeable future. If you need to build new infrastructure anyway, I’d recommend building it on top of DCE; you’ll be able to eliminate the separate storage network, and be able to choose between running FCoE and iSCSI over that network at any time.