In this first part of this face-off on FCoE network convergence, storage and networking expert Stuart Miniman says...
10 GbE and data center bridging are making it possible to implement FCoE now. Most will start with FCoE in the edge but soon move into the rack.
CIOs have a huge challenge in the face of explosive growth in data and applications: They must control not only the cost of IT infrastructure, but also deal with diminishing power and space availability. What does that mean for IT shops? It means the need to look at fundamental architectural changes, including converged networks and storage. Fibre Channel over Ethernet (FCoE) is finally ready to meet that challenge—first at the edge but very soon in the rack.
The IT community has worked for decades to deliver a single network for all infrastructures, but specific application requirements have spawned the development and adoption of multiple networks, including Ethernet for general networking, Fibre Channel (FC) as the primary storage network and InfiniBand for low latency and high-performance computing (HPC) environments.
FACE-OFF on FCoE
Stephen Foskett argues against this point of view and claims that end-to-end FCoE isn't ready for primetime, as it is unreliable and requires more engineering.
Yet this can change. With the general adoption of 10 Gigabit Ethernet (GbE) across enterprise data centers and the emergence of data center bridging (DCB), convergence to Ethernet as a single network is now viable. An added advantage is that DCB gives administrators the knobs to deliver Quality of Service (QoS) for all traffic—including general LAN traffic, plus storage options like iSCSI and FCoE—over a single network. While neither FC nor InfiniBand are likely to disappear in the next five years, new or expanded data centers should give strong consideration to converging on an all-Ethernet environment starting at the edge.
For FC customers, Fibre Channel over Ethernet is a path toward converged networks. The standard for FCoE was ratified in 2009 and a broad spectrum of products have been delivered— including those from Cisco, Brocade, Intel, HP, NetApp, EMC, Juniper, Dell and others. Adoption today is predominately in embedded solutions, such as blade servers, that are at the server-edge of the data center. This is the natural progression of technology adoption, especially in the risk-adverse storage world. FCoE at the edge is a simple deployment with a simple financial justification.
But FCoE won’t stop at the edge. Adoption is expected to increase in rack and stack servers. While servers and storage that support FCoE have been around for over a year, multi-hop FCoE configurations only started shipping a few months ago.
Multi-hop solutions face challenges in that despite conforming to the standard, architectures vary greatly among vendors. Yet this does not mean that there are stability issues; rather it means that some of the same switch-to-switch interoperability challenges seen in FC will carry over to Fibre Channel over Ethernet. FC customers typically standardize on a single vendor, and FCoE switches do have broad interoperability support with host adapters, storage arrays and FC switches.
While no storage administrator clamors for a new protocol, there is no need to fear FCoE, which is built with much of the same architecture as FC. As more of a company’s switches and adapters are “convergence-ready”—both Ethernet ports that can support FCoE and FC ports that can change personality to Ethernet/FCoE—there will be increased pressure from management to move toward a single network. The good news with Fibre Channel over Ethernet is that storage knowledge doesn’t go away with a move to an all-Ethernet environment.
In the second part of this face-off, Stephen Foskett argues that end-to-end FCoE for network convergence won't work.
About the Author: Stuart Miniman is an analyst and research lead for networking and virtualization for The Wikibon Project. He has worked extensively on supporting the adoption of new networking technologies to the market including Fibre Channel, iSCSI, FCoE and Distance Solutions (GigE, FCIP, iFCP, SONET, WDM). Stuart worked at EMC for 10 years; with a primary focus on storage networking and virtualization technologies.