Despite the longstanding goal of convergence and series of innovations designed to achieve it, the typical, contemporary data center remains a hodgepodge of heterogeneous Layer 2 protocols and standards. It is likely to stay that way for the foreseeable future.
Still, the various protocol innovations that have been developed in the pursuit of convergence – Fibre Channel over Ethernet (FCoE), 8G Fibre Channel, InfiniBand and Internet Small Computer Systems Interface (iSCSI), among them – stand to continue to play valuable roles in the storage and local area networks (SANs and LANs) of enterprises moving forward. And it is true that one emerging, high-end protocol might eventually shoulder all of the enterprise traffic.
Until that occurs, however, protocol-agnostic Wavelength Division Multiplexing (WDM) will continue to enable enterprise data-center and carrier service managers to adopt these disparate valuable protocols as new demands, capabilities and opportunities emerge.
There’s a swirl of industry interest today in FCoE, the latest darling in the search for an omnibus convergence fabric in enterprise networking.
We appear to be headed toward an industry-standard FCoE protocol (the open version is known as “Converged Enhanced Ethernet,” and some vendors use proprietary versions of different names) that would be designed to cost-effectively converge Fibre Channel and Ethernet. That is appealing news for enterprise data-center managers because these are the two most widely deployed protocols in enterprise networking. Consolidating LAN and SAN would significantly cut management and maintenance burden, as well as costs for servers, adapters, cables, etc.
FCoE would specify enhancements to conventional Ethernet in a number of key areas including flow control, packet loss and Quality of Service (QoS). As the technology matures to deliver enterprise-grade performance, reliability and quality characteristics, FCoE figures to have a terrific future in some important areas of enterprise networking:
- Initial-phase I/O consolidation
- In cases where an enterprise needs to move a lot of raw data over a network for storage backup, without concern for synchronous disaster-recovery or business-continuity services
- Greenfield opportunities where FCoE might possibly provide reliable, high-bandwidth transport without requiring investment in a Fibre Channel SAN
- Large, Internet-scale data centers – such as social-networking sites and search engines – and other environments that operate on a commodity model where low-cost connectivity is more important than reliability
- Analytics and high-performance computing
Enterprises, however, must beware of hype that touts FCoE as an immediate, one-size-fits-all convergence fabric. Some proposed, non-standard FCoE implementations, at least, fail to meet the levels of latency, synchronous recovery and continuous availability that some of today’s most important data-center services demand. Plus, in many cases, converging SAN and LAN via FCoE would require a costly and disruptive upgrade of the existing enterprise core with new low-latency Ethernet switches.
The more likely foreseeable role for FCOE, once it is ratified, is as a valuable enabler to specific, appropriate business applications – carried on the same WDM infrastructure between data centers with a variety of other services.
8G Fibre Channel
Another interconnect contender in the news of late is 8G Fibre Channel. Earlier this year, IBM announced it had established an end-to-end infrastructure based on the emergent technology. Meanwhile, data, voice and managed enterprise services provider COLT said it had achieved 8G Fibre Channel storage delivery over distances exceeding 135 kilometers.
Why all the activity around 8G Fibre Channel?
It’s a unique case of economics in this technology’s favor. A single 8G Fibre Channel link delivers as much bandwidth as four 2G Fibre Channel connections while requiring no additional equipment, port space, power or maintenance. So, the effect of an upgrade to 8G Fibre Channel is four times more bandwidth at significantly reduced cost when compared with the incremental addition of 2G Fibre Channel links to achieve the same capacity. In reducing the directors, servers, storage arrays, equipment racks and optical channels to be deployed and maintained, the typical data center would save hundreds of thousands of dollars in the first year alone. Plus, there are more savings to be realized because fewer devices have to be housed, powered and cooled with 8G Fibre Channel.
The enticing business case for 8G Fibre Channel puts it on the short list of protocols for high-bandwidth transport of SAN and LAN applications across the enterprise WDM infrastructure. In addition, carriers see 8G Fibre Channel as a key technology for enabling “Cloud” computing and high-end application, storage and compute services for enterprise customers who wish to concentrate costs on their core businesses and offload Information Technology (IT) responsibilities.
InfiniBand is one of those technologies that entered enterprise networking ballyhooed as the single, powerful, unifying protocol. While that has not occurred, InfiniBand has experienced intensified adoption in sync with increased enterprise reliance on latency-intolerant, high-bandwidth, synchronous applications such as high-end business continuity, disaster recovery, clustering and grid computing.
In this first decade of the 2000s, no application in enterprise networking has received more attention and, perhaps, investment than storage. Sophisticated business continuity and disaster recovery capabilities have been put into place by enterprises of almost every size across almost every industry. In some cases, the business driver was the threat of revenue losses associated with network downtime. In others, stiffened regulations around data loss and recovery came into play.
Similarly, clustering and grid computing continue to evolve from limited usage in the research-and-education community and other small niches of enterprise networking to wider reliance across industries. One of the benefits of these applications is that they enable an enterprise to realize significant cost advantages as workload is distributed across a geographic cluster of industry-standard servers as opposed to concentrated on fewer premium servers in one location. In the same way, overall system resiliency is improved and risk is reduced.
InfiniBand cost-effectively delivers the high bandwidth and low latency that these high-end applications require, and the technology enabling the protocol is firmly proven and widely available. Consequently, InfiniBand’s deployment across WDM networks has steadily grown as a key enabler for business activities as diverse as real-time stock trading and medical-image analysis. That means InfiniBand is increasingly supporting the applications that serve as the lifeblood to enterprises’ critical business operations. With InfiniBand uniquely delivering the networking characteristics that these applications require, the protocol is likely to play a critical role in the data center for years to come.
Two trends have facilitated iSCSI’s emergence as another of the key threads in the SAN fabric. First, small and medium-sized businesses have joined the Fortune 500 in adopting distributed storage strategies. Second, large companies have sought the cost advantages of a tiered storage strategy. In this approach, higher-end storage protocols and media are committed to mission-critical and runtime-sensitive traffic, while cost-effective and simple iSCSI accomplishes other objectives, such as backing up low-priority data over long distances.
iSCSI’s roles are defined, in part, by the header and commands the protocol must introduce to Ethernet packets being carried across the enterprise network. Data-center managers must factor in this overhead when determining where and for what applications to use iSCSI. Synchronous applications such as Geographically Dispersed Parallel Sysplex, Fiber Connection (FICON) and Enterprise Systems Connectivity (ESCON) are latency sensitive and bandwidth intensive (several terabits per second); higher-performance protocols are required.
But IP-based iSCSI creates valuable options for cost-conscious enterprises to better utilize Ethernet networks and interfaces that are widely deployed across the enterprise and familiar to IT support staff. Consequently, iSCSI is becoming a more prevalent solution for remote backup in specific areas of the network, with WDM supplying the reliable, transparent and seamless connectivity among SAN islands of iSCSI and other protocols.
Beyond the various technological and cost considerations, matters of enterprise behavior and organization will also play a role in keeping a mix of specialized protocols in play in the data center. In most enterprises today, separate organizations manage the SAN and LAN. Total convergence on any single interconnect means the storage and server groups must entrust their mission-critical traffic (and careers) to the network group. This is no small transformation, and it isn’t going to happen quickly.
For the foreseeable future, then, enterprises will continue to strategically weigh the technical requirements and business objectives of applications against the capabilities and costs of enabling protocols—and deploy FCoE, 8G Fibre Channel, InfiniBand, iSCSI and others appropriately (Figure 1).
This means WDM will continue to serve as the unifying force in enterprise networking. Agnostic to the protocols and bit rates being carried across optical networks, WDM consolidates the spectrum of SAN and LAN protocols across a powerful yet manageable infrastructure with the flexibility to accommodate emerging protocol innovations.
Todd Bundy is director global alliance management enterprise with ADVA Optical Networking (www.advaoptical.com). He has 26 years of experience in the storage networking industry, is a recognized expert in SAN and optical networking, and specializes in storage applications over various types of networks to meet corporate contingency plans.