Fibre Channel over Ethernet

Fibre Channel over Ethernet ( FCoE ) is a protocol for transmission of Fibre Channel frames in full duplex Ethernet -based networks. The main objective in introducing FCoE is used for I / O consolidation on the basis of Ethernet (IEEE 802.3) with regard to the reduction of physical complexity of network structures especially in data centers.

With FCoE, it is possible a unified physical infrastructure to use both for the transmission of Fibre Channel as well as conventional Ethernet. The scalability and higher bandwidth Ethernet-based set network structures with the currently popular 10 GBps ( soon also 40 or 100 Gbps ) key advantages delivered through the use of Ethernet come for transporting Fibre Channel frames on the other hand also disadvantages classical Ethernet protocol, such as frame loss in case of overload situations for supporting, so that some improvement to the Ethernet standard is necessary to ensure reliable transmission based on Ethernet. These extensions are driven by the term Data Center Bridging.

The standardization of FCoE began in April 2007 within the FC -BB -5 working group of T11 and on June 4, 2009, the INCITS passed for publication as FC -BB -5 draft standard.

Application and motivation

The main application of FCoE can be seen in the I / O consolidation ( also referred to as network convergence ). In this regard, the consolidation of Ethernet -based networks and storage area networks ( SAN) to mention a common, efficient physical infrastructure. The main advantages of this approach are likely to significantly lower the cabling and lower overall capital cost of a common infrastructure, as well as better resource utilization of physical infrastructure represent.

Another advantage is certainly seen in the currently prevailing virtualization strategy of many data center providers in interaction with FCoE, because ultimately FCoE in practice is also a kind of virtualization technology based on physical media, and as such partly to the host systems can be transferred for virtualized server. Such consolidation strategies can therefore:

• Reduce effort and cost of a physical infrastructure consisting of network elements and cables
• Such as servers reduce the number and total cost of the necessary network interface cards ( NICs) in terminals
• Reduce cost of operation ( energy supply and heat dissipation ).

Demarcation

FCoE transfers similar to iSCSI block data on minor protocol layers, however, the two protocols are fundamentally different. iSCSI can be classified as an application protocol according to the OSI model, as it uses TCP as the transport protocol. Thus, it is a routable protocol and therefore can be applied in principle beyond the boundaries of a data center out. With iSCSI, you take but because of these properties and the lack of additional requirements under higher layers a relatively large packet overhead into account, so that the total transport capacity is rather low classify in particular within a data center compared to Fibre Channel. FCoE, however, dispense with the use of network and transport layer protocols such as IP and TCP, and is equal to Ethernet. Here FCoE benefited from the significantly lower packet overhead and the associated increase in transport capacity and uses the local limit of Layer 2 domains in purchasing. In addition, special challenges for the used Ethernet standard, which is supported by Data Center Bridging bill. To achieve this requires a data center network infrastructure, which requires a migration generally difficult. Therefore, FCoE has a special significance in the case of construction of new power structures.

Functionality

FCoE encapsulates Fibre Channel natively in Ethernet frames with the FCoE protocol specifications, the layers FC -0 and FC -1 of the Fibre Channel stack replace ( see frame format). Thus, it is possible to integrate the Ethernet infrastructure without major changes existing in a storage area network ( new ). Because FCoE builds directly on the Ethernet protocol stack, it is significantly different from iSCSI, that also transmits SCSI block data over network protocols, but it is based on the TCP stack and therefore in contrast to FCoE is routable.

The standard of classical Ethernet originated at a time, in the through networks only manageable amounts of data have been transferred. Thus, in this period there was little need for flow control mechanisms ( engl. flow control ). Fibre Channel on the other hand implements flow control mechanisms, since block data A., are very sensitive to transmission errors and can not otherwise be intercepted. For this reason, FCoE requires enhancements to the classic Ethernet standards to Flusssteuerungsmechnismen to prevent overload situations and the associated frame losses. In these extensions, the IEEE is working in the Data Center Bridging Task Group.

The major enhancements to the Ethernet standard in Data Center Bridging can be summarized as follows:

• Priority -based Flow Control ( PFC) Version 0 Specification ( Submitted to IEEE 802.1Qbb Working Group ).
• Enhanced Transmission Selection (ETS) Version 0 Specification ( Submitted to IEEE 802.1Qaz Working Group ).

Furthermore, it must meet the following requirements Ethernet:

• Encapsulation of native Fibre Channel frames in Ethernet frames.
• A mapping between Fibre Channel N port identifiers ( FCIDs ) and Ethernet MAC addresses.

Terminal adapter

Terminals are connected directly with the help of so-called converged network adapters ( CNAs ) with the FCoE fabric. Such an adapter provides both Fibre Channel host bus adapter ( HBA ) as well as classic network interface card functions on a hardware. This relieves the CPU at low level frame processing and provision classical SCSI functionality. FCoE encapsulation may alternatively be carried out in software, sometimes under significant CPU usage.

The first FCoE device was implemented by Frederick Knight for NetApp.

Frame format

Both Fibre Channel and classic networks have their own stacks, each providing a set of functionalities. FC stack consists of five layers ( FC -0 to FC- 4), while Ethernet is typically represented in a 7-layer OSI reference model, where it covers the first two layers (physical layer and data link layer ). FCoE is designed to transmit the FC - FC - 2 layer in the Ethernet. Thus the upper Fibre Channel layers FC -3 and FC -4 can finally be imaged via IEEE 802.3 Ethernet layers. A typical FC frame has a data length of up to 2112 bytes plus header and CRC. The maximum size of an FCoE frame is 2180 bytes. Thus FCoE requires the support of Ethernet infrastructure for the transmission of the so-called baby jumbo frames with a size of up to 2.5 KB, so the FC frames contained will not be split.

Specifying the first 48 bits of the frame, the destination MAC address, the second 48 bits, however, the source MAC address. The 32-bit IEEE 802.1Q tag provides the same functionality as for the VLANs and provides multiple virtual local area networks on the same physical infrastructure. FCoE is encapsulated directly in Ethernet frames using a specific Ethertype ( 0x8906 ). For an additional 16 bits are provided, followed by a 4 -bit version field. The next 100 bits are reserved and another 8 bit index the start of the Fibre Channel frame. After an 8-bit end-of -frame delimiter (EOF ) followed by further 24 reserved bits. The frame ends with 32 bits for the FCS function, which provides error detection for the Ethernet frame.

The encapsulated Fibre Channel follows the Ethernet header, as in the actual FC standard, a 24 -byte header, followed in turn by FC FC payload, including Fibre Channel CRC. The usual cyclic redundancy check is ( engl. cyclic redundancy check, therefore, usually CRC) used for error correction. As the classic FC frame remains completely intact, and continues to be used, it is possible to connect an existing Fibre Channel SAN with an FCoE switch, the FCoE switch then acts simply as a gateway and removes the outer Ethernet frame, the result is an integration of functions in accordance with FC - Ethernet, without the need for a dedicated gateway.

Another important component of the FCoE FCoE Initialization Protocol is the standard (FIP ) that the FCoE capabilities of the network components of an Ethernet cloud detects and initializes. FIP turn uses a dedicated Ethertype ( 0x8914 ).

Similar standards

• ATA over Ethernet: For ATA over Ethernet ( ATAoE ) are encapsulated ATA / ATAPI packets into Ethernet. Similar to FCoE ATAoE is not encapsulated in TCP / IP, ATAoE is not routable.
• Fibre Channel over IP Fibre Channel over IP ( FCoIP ) Fibre Channel packets are encapsulated in TCP / IP. FCoIP is therefore routable.
• HyperSCSI In HyperSCSI SCSI packets are encapsulated in Ethernet. Similar to FCoE HyperSCSI is not encapsulated in TCP / IP, HyperSCSI is not routable.
• ISCSI ( SCSI over IP): With iSCSI, SCSI packets are encapsulated in TCP / IP. This results in a higher overhead, but this is iSCSI routable and can be as Fibre Channel over IP, used in wide area networks.