Why FCoE is Not the Solution

Fibre Channel over Ethernet (FCoE) by itself is not a truly significant leap forward in data center I/O. Under the guise of innovation, FCoE hides the same old I/O architecture. Let me explain. Some contend that FCoE solves the I/O complexity problem by carrying block storage and networking traffic on the same wire. Others claim that FCoE represents a grander vision, promising to simplify the management of data center I/O by providing a single-network solution that runs across all devices. The reality is that FCoE solves neither the complexity nor the management problems. It is a minor change to the status quo when a major leap forward comparable to server virtualization is needed for I/O.

Let’s start with the interconnect – NICs/HBAs, cables and switches. FCoE is a technology which allows Fibre Channel traffic to travel over enhanced Ethernet rather than over native Fibre Channel. The idea is to use special (and costly!) converged NICs (CNAs) which integrate the NIC and HBA functionality, along with special switches which support an enhanced version of Ethernet combined with FC services. This makes it possible for FC block storage traffic as well as regular network traffic to be carried over a single Ethernet network. The end result is that the number of interconnect elements is cut in half, which is why FCoE is a change of sorts to the status quo, but the problem is that this change is not fundamental – it begins and ends with cutting the interconnect in half. Everything else stays the same, including the complexity, rigidity, and management difficulty of I/O.

What you need to manage in a flexible and consolidated data center are the endpoints of the I/O interconnect – the properties of the NICs and HBAs within the servers, and the connections of the servers to the available Ethernet and Fibre Channel networks. The I/O identity of a server is determined by its I/O devices and the physical networks to which these devices are connected. This is true regardless of whether the interconnect is FCoE or separate Ethernet and Fibre Channel. The fact is that FCoE provides no help whatsoever in managing the I/O identity of servers, and it doesn’t even deliver an improvement in network and SAN management since those burdens remain the same as they were before, or even worse due to the complexity of mapping FC concepts such as VSANs to Ethernet concepts such as VLANs.

With FCoE, you still have I/O devices (CNAs) within the server with properties which are not centrally managed, and with the limitations of physical wiring. The physical connectivity of servers still cannot be controlled other than by physical intervention. For example, if you want to change the connectivity of a server from one Ethernet network to another, you still cannot do it since that connectivity is physically determined by the wire coming out of the CNA and the Ethernet port to which it is connected. If you wish to connect a server to multiple distinct Ethernet networks, you still need to use multiple CNAs, cables, and switch ports. If you want a new server to assume the identity of an old server, including MAC addresses, WWNs, physical connectivity, and boot image, FCoE by itself will not provide the solution.

Solving the complexity and management problems of I/O requires a transition from physical I/O to virtual I/O. NICs and HBAs must become virtual NICs (vNICs) and virtual HBAs (vHBAs) which can be managed centrally. The physical connectivity of vNICs and vHBAs to Ethernet and Fibre Channel networks needs to become the responsibility of an external device which can change this connectivity on-demand. vNIC and vHBA traffic is carried on a single fabric which is now truly converged. Through this fabric, servers can connect to any Ethernet or Fibre channel networks as needed without physical intervention. It’s the flexibility and central management of the properties of the virtual I/O resources and their connectivity to data center networks that matters most. It is the transition from physical I/O to virtual I/O – not the mere convergence of storage and networking on the same wire – which will provide the real leap forward in cost and complexity reduction as well as better management and control of data center I/O.

Tags:

4 Responses to “Why FCoE is Not the Solution”

  1. [...] I like to argue. I’ve been wishing for an argument on FCoE for some time now (a wish in which I’m not alone), but all I run across is how FCoE is going to save the [...]

  2. [...] recently came across this blog entry over at Xsigo’s new corporate blog, I/O Unplugged. A key phrase in this blog entry really [...]

  3. [...] recently came across this blog entry over at Xsigo’s new corporate blog, I/O Unplugged. A key phrase in this blog entry really caught [...]

  4. [...] Why FCoE is Not the Solution « Xsigo Virtual I/​O Blog —  [...]

Leave a Reply