Converged networks with Fibre Channel over Ethernet and Data Center Bridging

8
TCs, the existing mechanisms can cause Quality of Service (QoS) issues, limit the ability to scale a network,
and affect performance.
DCB standards
DCB is not just the name for a set of new standards the IEEE is developing. It is a term often used for
Ethernet designed to carry multiple TCs, some with lossless behavior. You can think of DCB-enabled
Ethernet as applying the DCB standards to IEEE 802.3 Ethernet standards to create a new set of products to
implement this improved version of Ethernet. The change from legacy Ethernet to DCB-enabled Ethernet
requires hardware and software changes, so you can’t upgrade legacy Ethernet NICs and switches with
DCB support to carry FCoE traffic. Fortunately, you only have to update the data paths in a data center that
carry FCoE with DCB-enabled Ethernet devices.
For full end-to-end data center use, all equipment manufacturers must agree to adopt four new IEEE
protocols. The proposed standards are still under development, and full ratification of the complete set may
take until late 2010 or 2011. One result of these ongoing standardization efforts is that DCB/FCoE
products offered on the market today will likely need frequent software upgrades or even new hardware by
the time DCB/FCoE technology is fully mature.
The DCB Task Group within the IEEE 802.1 Higher Layer LAN Protocols Work Group is defining DCB for
protocols and technologies that apply to data center-oriented LAN communications. The standards they
develop apply to all IEEE 802 network types, but they implicitly target Ethernet for primary implementation.
Table 1 lists four new technologies defined in three DCB draft standards.
Table 1. DCB draft standards for IEEE 802 networks
Draft standard New technology
IEEE 802.1 Qbb Priority-based Flow Control (PFC)
IEEE 802.1 Qaz Enhanced Transmission Selection (ETS)
DCB Capability Exchange Protocol (DCBX)
IEEE 802.1Qau Quantized Congestion Notification (QCN)
These standards serve three general purposes:
• Allow IEEE 802 LANs to carry multiple traffic classes
• Support lossless behavior on a subset of these traffic classes
• Formally define standard frame transmission scheduling mechanisms to support multiple traffic classes.
You don’t have to use all four of these protocols to implement a DCB network, and you don’t need to use all
options available in each protocol. However, if vendors do not implement the entire set, products may limit
the possible scale or features. Because the standards are evolving, current DCB/FCoE products do not
implement all of these protocols or all their supported options. Therefore, we must discuss their deployment
limitations within a data center.
Priority-based Flow Control
Legacy FC networks support a link-level flow control mechanism known as buffer-to-buffer or credit-based
flow control. This lightweight, high performance mechanism lets FC work in a lossless manner. Credit-based
flow control provides a reliable layer 2 network required for block storage traffic, for example SCSI. To
transport FC and SCSI protocols over Ethernet and maintain a lightweight implementation, we recommend
providing a similar mechanism for Ethernet networks.
Legacy Ethernet uses a simple flow control mechanism. It uses pause frames to let a congested network
device port on an Ethernet NIC or switch tell its link partner to pause all traffic for a specified time. This
approach can limit performance when a network device port has multiple queues for receiving incoming