FCoE “and” iSCSI - Who cares? It’s all about Data Center Ethernet
Innovation in networking has always been associated with animated (sometimes religious, but who cares) debates about what’s right and what’s wrong. And it was only a matter of hours before the iSCSI vs. FCoE debate got started.
This time, though, I do not believe it is a either/or type of situation, but rather an “and”...
Marc Farley is a brilliant storage expert, who I have had the pleasure to work with in one of my past life. I always enjoy his insights into technology and I love the fact that he believes in what he writes and he does it with passion. Marc posted an entry today on “FCoE is a great dead end”. Considering some of the positions he has already taken in the past on FCoE and iSCSI, I was not surprised to see that, but I think his passion came out stronger than some of the technical arguments he made. I did not completely disagree with his statements, but I think some of them would require further clarification and are probably less black and white than they seem.
A follow on post by Matt Baker on “FCoE, let’s not be too hasty…” is much closer to my view of the storage networking world and I would overall agree with it, even though I am less optimistic on the universal applicability of iSCSI and I would point out that FCoE is as available as iSCSI today.
I certainly agree that Data Center Ethernet (if properly implemented) is the real key differentiator and enabler of Unified Fabric, whether we like to build it with iSCSI or FCoE. Nonetheless, as much as FCoE and iSCSI may sound similar in the objective they try to achieve, they are substantially different in the way they do it, and because of this they may be applicable to different environments.
FCoE is not a new storage protocol, it is just a new transport for Fibre Channel; Fibre Channel sees an FCoE link as a different type of physical media and nothing else. iSCSI looses any relationship with Fibre Channel (some may claim this is a good thing…) and is a new transport for SCSI, which implies a more substantial change to the storage stack.
It is true that FCoE is a “stupid” technology (if simple can be synonim of stupidity,) but its “stupidity” is its strength: it does not change anything in Fibre Channel, it is Fibre Channel over a better and faster media. All the key concept of Fibre Channel are preserved intact: WWNs, Zoning, DNS, RSCN, FSPF, etc work unmodified. iSCSI does all of those things in a much “better” way, but it’s not the same way as Fibre Channel. I would completely agree that iSCSI is a “superior” protocol to FCoE. iSCSI is the absolutely correct way to define a transport for SCSI over IP, but still that deos not make iSCSI easily applicable in environments (especially in the enterprise) where Fibre Channel is the prevalent storage transport and backward compatibility and co-existence are a key requirement.
iSCSI is perfect for environments where there is no FIbre Channel legacy and maybe there is no existing IT competence around Fibre Channel. Small-Medium Businesses will greatly benefit of iSCSI and possibly also some larger environments where storage consolidation has not happened, yet.
FCoE, on the other end, allows for an easier coexistence between Fibre Channel and Ethernet-based storage networking, and it represents an easier and smaller step towards Unified Fabric. In a hybrid FCoE/FC environment there is no need for stateful gateways, while in a hybrid iSCSI/FC environment you need to have a stateful gateway, which will necessarily introduce additional latency and complexity (we ship these gateways, so I am not being exactly politically correct either, here.)
The bottom line is that I agree with Matt and I think that both FCoE and iSCSI have a role in evolving storage networks and ultimately making storage much more pervasive. The key enabler for both is Data Center Ethernet and the ability to leverage a lossless transport for storage traffic. To this last point, I would like to point out that TCP delivers better performance over a lossless transport, and I am willing to bet that it will not be long before we see iSCSI over UDP with Data Center Ethernet. That would be an interesting development…
Posted by Dante Malagrino at 11:58PM PST
MarcFarley Apr 11, 2008
Hi Dante,
Thanks for your input here. We probably agree more than we disagree. My guess is that the customers we come into contact with tend to be different types. I tend to spend more time with medium sized business customers than the Fortune 1000.
iSCSI over UDP… I thought that was already tried but rejected. I know there were several attempts ‘in the day” to do it. SOIP and EtherSCSI. Neither got very far.