In the first part of this face-off on FCoE network convergence, Stuart Miniman discussed FCoE starting at the edge then moving into the rack. In this second part, storage and networking expert, Stephen Foskett argues that FCoE at the edge might be a step toward network convergence, but end-to-end FCoE needs a lot more engineering and isn't reliable.
All last year I watched storage and networking vendors twisting and turning to convince the world they have delivered an end-to-end FCoE protocol. In fact, the products they released are largely inadequate. What’s more, they’ve never addressed the burning question: Why bother with an end-to-end strategy for network convergence when the FCoE protocol at the edge is more practical?
It’s important to first note that moving enterprise storage traffic to Ethernet networks seems like a match made in hell. The SCSI protocol requires delivery of packets that are lossless and in order, but Ethernet was designed for “best effort” delivery. This won’t cut it for storage, which is a high-volume payload, swamping adapters and switches with I/O.
FACE-OFF on FCoE
Stuart Miniman disagrees with this point of view, saying that FCoE network convergence is ready for primetime, as 10 GbE and data center bridging are emerging.
But the lure of commodity-driven pricing and order of magnitude faster roadmap performance is too enticing to ignore, so the storage networking industry has left it to the engineers to figure out how to make Ethernet an appropriate transport mechanism for block storage.
Fitting the square peg of storage into the round hole of Ethernet required quite a bit of engineering:
The FCoE protocol frame format and FCoE Initialization Protocol (FIP), established in 2010 as FC-BB-5, laid the groundwork, while data center bridging (DCB) extensions brought flow control and queue management to transform Ethernet into a reliable transport mechanism.
This bulked up version of 10 Gigabit Ethernet (GbE) functions surprisingly well in practice, and was quickly put into production at the edges of existing Fibre Channel SANs for blade server attachments. This is the essence of “edge-only” FCoE, and it delivers a one-two punch of flexibility and performance at a reasonable cost. Most large IT shops are perfectly happy using Fibre Channel at the core and Ethernet at the edge.
Most large IT shops are perfectly happy using Fibre Channel at the core and Ethernet at the edge.
But network switching vendors won’t be content until they convert the whole SAN to Ethernet, so they spent 2011 crowing about end-to-end FCoE, even though products that shipped are a mixed bag of pre-standard and proprietary technologies. Their Ethernet fabric approaches range from functional-but-funky to standardish-but-experimental to laughably-limited. And implementation of FC-BB-5 is decidedly spotty for most vendors. Put simply, end-to-end FCoE is premature.
Ultimately, end users will be happier deploying 8 Gb Fibre Channel SANs with a mix of FC and FCoE server connections. They can see how shaky FCoE at the core is at the present time, and they are perfectly happy holding off on that transition for a few more years. Why would they risk their jobs, and the safety of their data, for a brand-new protocol with limited return on investment?
Perhaps this controversy is born of a fundamental misunderstanding by the networking industry of the nature of enterprise storage. “Storage people” are cautious and risk averse. Adoption of new technologies is slow because storage simply must be reliable. FCoE protocol proponents should be pleased with their foothold at the edge of the SAN rather than pushing aggressively to the core.
About the Author: Stephen Foskett is an independent consultant and author specializing in enterprise storage and cloud computing. He is responsible for Gestalt IT, a community of independent IT thought leaders, and organizes their Tech Field Day events. He can be found online at GestaltIT.com, FoskettS.net, and on Twitter at @SFoskett.