04-15-2011 10:41 AM
So, like most things in the competitive business environment, once the Storage Area Network, (SAN), had established itself as the dominant IT server to storage infrastructure, based on the Fibre Channel , (FC) network architecture, there were parts of the IT industry that were arguing, not with the value of a SAN, but whether or not it needed to be a FC SAN. An early argument centered on iSCSI, which is essentially the porting of the SCSI protocol to run on an IP network, which to most folks meant an Ethernet network. iSCSI as a solution, eventually evolved over time to occupy a lower rung on the SAN architectural ladder, as in some cases it means a less expensive SAN infrastructure, which can fit with certain requirements.
However, the industry is now revisiting this issue in the form of the FCoE protocol, which is a standard specification, defined in the ANSI T11.3 subcommittee as part of the FC-BB-5 Standard, that defines how the entire FC frame, can be encapsulted inside an Ethernet frame, and put onto an Ethernet network. The FC-BB-5 standard is actually a Fibre Channel Backbone standard and covers many use cases for connecting FC source and destination devices. The FC-BB-5 standard defines FCoE in a section known as FC-BB_E, which defines that the Fibre Channel frames are tob be transported, over a lossless Ethernet network. So it's a new protocol, for a new Ethernet, since we know that Ethernet is a lossy, not lossless, network architecture. The benefits of moving FC traffic over an Ethernet infrastructure are described as including simplified network management, elimination of redundant cabling, switches, etc., reduced power and heat requirements, and enhanced data network performance, since FCoE is built on the new 10 Gbps, lossless, "enhanced" Ethernet, embodied in the 802.3 working group set of standards, generically referred to as the DCB, (Data Center Bridging), standards.
This has lead to a number of prognostications that the evolving data center network architecture will eventually become one converged, onto a DCB Ethernet network.
The problem is that many of these perceived benefits are projected, since both the protocol, and network architecture are brand new. Noone can definitively say how DCB/FCoE networks will perform, or what benefits they can deliver, relative to many running FC networks, until you can build them to carry the same kind of traffic, under the same kinds of conditions. Test cases for this, are still a ways off. The industry is still working to develop standards for the technologies that they feel will be required for this new architecture, at the same time that vendors are developing and building products, which will support these new standards.
It's ironic to note this effort to re-architect Ethernet, will actually make it operate more like Fibre Channel operates today, (lossless, low latency, etc.). So it begs the question, that if we need to re-architect the network in our data center, should we start by dismantling the part that Ethernet is trying to be more like? Or perhaps should we look at re-architecting the current Ethernet network instead. One of the biggest selling points on the concept of converged networks is reducing cabling, and related network components. The problem with this math is that the excess network cabling and switches isn't in the FC network, it's in the Ethernet network.
Certainly one of the most interesting aspects to this question, is the impact of DCB networks on current Ethernet/TCP/IP architectures. The new DCB Ethernet looks like it will be able to deliver Ethernet architects a more flexible, powerful, TCP/IP network, than the currently classic Ethernet. Additionally, in Brocade's DCB Ethernet solution, the VCS Ethernet Fabric architecture, with a re-design of Ethernet, comes the opportunity to include new technologies in the new Ethernet network architecture which would be ideally suited to supporting Virtual Machine, (VM), environments, such as the new Automatic Migration of Port Profiles, (AMPP).
As far as the Fibre Channel technology is concerned, this year will see the introduction of Brocade's seventh-generation SAN technology, built on a new FC ASIC, code named, Condor3. The new Condor3 switches will include an new set of features which should greatly advance the scalability, availability, manageability, and performance of the FC network architecture, beyond what even exists today. (more on that later)
So it appears that the future is bright for growing and extending the FC network architecture, while at the same time, Ethernet is being reinvented to operate more like FC, which will make the Ethernet/IP network more performant, lossless, and more "vm-aware". I don't know if you're experiencing an "either/or" scenario in your enterprise where a debate exists about converging onto a single network architecture like DCB, but it's clear that both FC and DCB network architectures have a lot to add to the data center, And that they can do so best, when deployed together.
What are your thoughts?