The Role of Data Center Bridging in LAN-SAN Convergence

A great deal of attention is being focused on the potential convergence of LANs and SANs in the data center. And it's more than just vendor hype. As recently reported in TechNotes, a number of companies intend to complete at least a moderate deployment of LAN-SAN convergence over the next two years. But just what will it take to bring that about?

Lessons from the VoIP Trenches

Certainly this isn't the first time two complementary technologies with significant differences have been converged. Take VoIP, for example. Looking at the hurdles that were overcome to bring data and voice together and make VoIP a marketplace reality can provide tremendous insight into the challenges that a converged data center LAN presents.

A major obstacle to early VoIP implementation was that the quality of a voice call is sensitive to delay, jitter and packet loss, and the data networks of that era lacked the mechanisms to minimize these metrics to ensure acceptable performance levels. In the years following VoIP's introduction, numerous techniques - among them sophisticated congestion control mechanisms - were introduced into data networks, enabling successful VoIP implementations by the vast majority of IT organizations.

A somewhat similar situation applies in LAN-SAN convergence. In particular, traditional Ethernet provides only a best-effort service that allows buffers to overflow during periods of congestion; it relies on upper-level protocols such as TCP to manage congestion and recover lost packets through retransmissions. In an integrated LAN-SAN, this could result in a level of delay that is unacceptable for storage functions.

To emulate the lossless behavior of a Fibre Channel (FC) SAN, Ethernet needs enhanced flow-control mechanisms that eliminate buffer overflows for high-priority traffic such as storage-access flows. Lossless Ethernet is based on a set of standards commonly referred to as IEEE Data Center Bridging (DCB).

DCB's Key Components

DCB has three key components. One is the IEEE 802.1Qbb Priority-based Flow Control (PFC) standard, which allows the creation of eight distinct virtual link types on a physical link, with each virtual link mapped to an 802.1p traffic class. Every virtual link can be allocated a minimum percentage of the physical link's bandwidth. Flows are controlled on each virtual link via a pause mechanism, which can be applied on a per-priority basis to prevent buffer overflow and eliminate packet loss due to link-level congestion.

The second key component of DCB is the IEEE 802.1Qau Congestion Notification (CN) standard. This standard focuses on traffic-management techniques that eliminate congestion by applying rate limiting or back pressure at the network edge to protect the upper network layers from buffer overflow. CN is intended to provide lossless operation in end-to-end networks that consist of multiple tiers of cascaded layer 2 switches - the architecture typically used in larger data centers to support server interconnect, cluster interconnect and extensive SAN fabrics.

The third component is the IEEE 802.1Qaz Enhanced Transmission Selection (ETS) standard, which specifies advanced algorithms for the allocation of bandwidth among traffic classes, including the priority classes supported by 802.1Qbb and 802.1Qau.

DCB lossless Ethernet will play a key role in supporting Fibre Channel over Ethernet (FCoE) technology, which will allow the installed base of Fibre Channel storage devices and SANs to be accessed by Ethernet-attached servers with converged FCoE network adapters over a unified data center switching fabric. DCB will benefit not only block-level storage, but also all other types of loss- and delay-sensitive traffic. In the storage arena, DCB will improve network-attached storage (NAS) performance and make Internet Small Computer System Interface (iSCSI) SANs based on 10/40/100 gigabit Ethernet a more competitive alternative to 2/4/8Gbps Fibre Channel SANs.

It Won't Happen Overnight

So, the good news is that techniques now being developed and implemented will likely enable the convergence of the LAN and SAN. The bad news is that, for a variety of technological and cultural reasons, convergence tends to take a long time. If that seems counter-intuitive, just think back to earlier this decade and remember how many years were identified as being "The Year of VoIP."