Ethernet represents an ideal candidate for I/O consolidation. Ethernet is a well-understood and widely deployed medium that has taken on many consolidation efforts already. Ethernet has been used to consolidate other transport technologies such as FDDI, Token Ring, ATM, and Frame Relay networking technologies. It is agnostic from an upper layer perspective in that IP, IPX, AppleTalk, and others have used Ethernet as transport. More recently, Ethernet and IP have been used to consolidate voice and data networks. From a financial aspect, there is a tremendous investment in Ethernet that also must be taken into account.
For all the positive characteristics of Ethernet, there are several drawbacks of looking to Ethernet as an I/O consolidation technology. Ethernet has traditionally not been a lossless transport and relied on other protocols to guarantee delivery. In addition, a large portion of Ethernet networks range in speed from 100 Mbps to 1 Gbps and are not equipped to deal with the higher-bandwidth applications such as storage.
New hardware and technology standards are emerging that will enable Ethernet to overcome these limitations and become the leading candidate for consolidation.
10-Gigabit Ethernet (10GbE) represents the next major speed transition for Ethernet technology. Like earlier transitions, 10GbE started as a technology reserved for backbone applications in the core of the network. New advances in optic and cabling technologies have made the price points for 10GbE attractive as a server access technology as well. The desire for 10GbE as a server access technology is driven by advances in computer technology in the way of multisocket/multicore, larger memory capacity, and virtualization technology. In some cases, 10GbE is a requirement simply for the amount of network throughput required for a device. In other cases, however, the economics associated with multiple 1-G ports versus a single 10GbE port might drive the consolidation alone. In addition, 10GbE becoming the de facto standard for LAN-on-motherboard implementations is driving this adoption.
In addition to enabling higher transmission speeds, current 10GbE offerings provide a suite of extensions to traditional Ethernet. These extensions are standardized within IEEE 802.1 Data Center Bridging. Data Center Bridging is an umbrella referring to a collection of specific standards within IEEE 802.1, which are as follows:
- Priority-based flow control (PFC; IEEE 802.1Qbb): One of the basic challenges associated with I/O consolidation is that different protocols place different requirements on the underlying transport. IP traffic is designed to operate in large wide area network (WAN) environments that are global in scale, and as such applies mechanisms at higher layers to account for packet loss, for example, Transmission Control Protocol (TCP). Because of the capabilities of the upper layer protocols, underlying transports can experience packet loss and in some cases even require some loss to operate in the most efficient manner. Storage area networks (SANs), on the other hand, are typically smaller in scale than WAN environments. These protocols typically provide no guaranteed delivery mechanisms within the protocol and instead rely solely on the underlying transport to be completely lossless. Ethernet networks traditionally do not provide this lossless behavior for a number of reasons including collisions, link errors, or most commonly congestion. Congestion can be avoided with the implementation of pause frames. When a receiving node begins to experience congestion, it transmits a pause frame to the transmitting station, notifying it to stop sending frames for a period of time. Although this link-level pause creates a lossless link, it does so at the expense of performance for protocols equipped to deal with it in a more elegant manner. PFC solves this problem by enabling a pause frame to be sent only for a given Class of Service (CoS) value. This per-priority pause enables LAN and SAN traffic to coexist on a single link between two devices.
- Enhanced transmission selection (ETS; IEEE 802.1Qaz): The move to multiple 1-Gbps connections is done primarily for two reasons:
- The aggregate throughput for a given connection exceeds 1 Gbps; this is straightforward but is not always the only reason that multiple 1-Gbps links are used.
- To provide a separation of traffic, guaranteeing that one class of traffic will not interfere with the functionality of other classes. ETS provides a way to allocate bandwidth for each traffic class across a shared link. Each class of traffic can be guaranteed some portion of the link, and if a particular class doesn’t use all the allocated bandwidth, that bandwidth can be shared with other classes.
- Congestion notification (IEEE 802.1Qau): Although PFC provides a mechanism for Ethernet to behave in a lossless manner, it is implemented on a hop-by-hop basis and provides no way for multihop implementations. 802.1Qau is currently proposed as a mechanism to provide end-to-end congestion management. Through the use of backward congestion notification (BCN) and quantized congestion notification (QCN), Ethernet networks can provide dynamic rate limiting similar to what TCP provides only at Layer 2.
- Data Center Bridging Capability Exchange Protocol extensions to LLDP (IEEE 802.1AB): To negotiate the extensions to Ethernet on a specific connection and to ensure backward compatibility with legacy Ethernet networks, a negotiation protocol is required. Data Center Bridging Capability Exchange (DCBX) represents an extension to the industry standard Link Layer Discovery Protocol (LLDP). Using DCBX, two network devices can negotiate the support for PFC, ETS, and Congestion Management.
Fibre Channel over Ethernet
Fibre Channel over Ethernet (FCoE) represents the latest in standards-based I/O consolidation technologies. FCoE was approved within the FC-BB-5 working group of INCITS (formerly ANSI) T11. The beauty of FCoE is in its simplicity. As the name implies, FCoE is a mechanism that takes Fibre Channel (FC) frames and encapsulates them into an Ethernet. This simplicity enables for the existing skillsets and tools to be leveraged while reaping the benefits of a Unified I/O for LAN and SAN traffic.
FCoE provides two protocols to achieve Unified I/O:
- FCoE: The data plane protocol that encapsulates FC frames into an Ethernet header.
- FCoE Initialization Protocol (FIP): A control plane protocol that manages the login/logout process to the FC fabric.
Figure 8-1 provides a visual representation of FCoE.
Figure 8-1. Fibre Channel over Ethernet
When Fibre Channel frames are encapsulated in an Ethernet, the entire Fibre Channel frame, including the original Fibre Channel header, payload, and CRC are encapsulated in an Ethernet. Figure 8-2 depicts this.
Figure 8-2. Fibre Channel Frame Encapsulated in an Ethernet
The ANSI T11 specifies the frame format for FCoE. It is a standard Ethernet frame with a new EtherType of 0x8906. Also note that the new Ethernet frame has a new Frame Check Sequence (FCS) created rather than using the FCS from the Fibre Channel frame. Figure 8-3 illustrates the FCoE frame format.
Figure 8-3. FcoE Frame Format
FCoE standards also define several new port types:
- Virtual N_Port (VN_Port): An N_Port that operates over an Ethernet link. N_Ports, also referred to as Node Ports, are the ports on hosts or storage arrays used to connect to the FC fabric.
- Virtual F_Port (VF_Port): An F_port that operates over an Ethernet link. F_Ports are switch or director ports that connect to a node.
- Virtual E_Port (VE_Port): An E_Port that operates over an Ethernet link. E_Ports or Expansion ports are used to connect Fibre Channel switches together; when two E_Ports are connected the link, it is an interswitch link (ISL).
To facilitate using FCoE an additional control plane protocol was needed and thus FCoE Initialization Protocol (FIP) was developed. FIP helps the FCoE perform VLAN discovery, assists the device in login (FLOGI) to the fabric, and finds key resources such as Fibre Channel Forwarders (FCFs). FIP is its own Ethertype (0x8914), which makes it easier to identify on a network and helps FIP Snooping devices identify FCoE traffic. Figure 8-4 depicts where FIP starts and ends and where FCoE takes over.
Figure 8-4. FIP Process
FIP can be leveraged by native FCoE-aware devices to help provide security against concerns such as spoofing MAC addresses of end nodes and helps simpler switches, such as FIP Snooping devices, learn about FCoE traffic. This awareness can provide security and QoS mechanisms that protect FCoE traffic from other Ethernet traffic and can help ensure a good experience with FCoE without the need to have a full FCoE stack on the switch. Currently the Nexus 4000 is the only Nexus device that supports FIP snooping.
Single-Hop Fibre Channel over Ethernet
Single-hop FCoE refers to an environment in which FCoE is enabled on one part of the network, frequently at the edge between the server and the directly connected network switch or fabric extender. In a single-hop topology the directly connected switch usually has native Fibre Channel ports which in turn uplink into an existing SAN, although you can have a complete network without any other fibre channel switches. Single-hop FCoE is the most commonly deployed FCoE model because of its double benefit of seamless interoperability into an existing SAN and the cost savings with a reduction in adapters, cabling, and optics to servers.
This reduction in cabling and adapters is accomplished through the use of a new adapter: Converged Network Adapter (CNA). CNAs have the capability to encapsulate Fibre Channel frames into Ethernet and use a 10GbE Ethernet interface to transmit both native Ethernet/IP traffic and storage traffic to the directly connected network switch or fabric extender. The CNA’s drivers dictate how it appears to the underlying operating system, but in most cases it appears as a separate Ethernet card and separate Fibre Channel Host Bus Adapter (HBA).
Figure 8-5 shows how a CNA appears in Device Manager of a Microsoft Windows Server.
Figure 8-5. CNA in Device Manager
Using CNAs in a server, a typical single-hop FCoE topology would look like Figure 8-6 where a server is connected to Nexus 5x00 switches via Ethernet interfaces. The Nexus 5x00 switches have both Ethernet and native Fibre Channel interfaces for connectivity to the rest of the network topology. The fibre channel interfaces connect to native fibre channel ports on the Cisco MDS switches, and the Ethernet interfaces connect to the Ethernet interfaces on the Nexus 7000 switches. The FCoE traffic is transported only across the first or single hop from the server to the network switch. The current implementation of the Cisco Unified Computing System (UCS) uses single-hop FCoE between the UCS blade servers and the UCS Fabric Interconnects.
Figure 8-6. Single-Hop FCoE Network Topology
Multhop Fibre Channel over Ethernet
Building on the implementations of single-hop FCoE, multihop FCoE topologies can be created. As illustrated in Figure 8-6, native fibre channel links exist between the Nexus 5x00 and the Cisco MDS Fibre Channel switches, whereas separate Ethernet links interconnect the Nexus 5x00 and Nexus 7000. With multihop FCoE, topologies can be created where the native fibre channel links are not needed, and both fibre channel and Ethernet traffic use Ethernet interfaces.
The benefit of multihop FCoE is to simplify the topology and reduce the number of native fibre channel ports required in the network as a whole. Multihop FCoE takes the same principles of encapsulating fibre channel frames in Ethernet and uses it for switch-to-switch connections, referred to as Inter-Switch Links (ISL) in the Fibre Channel world, and uses the VE port capability in the switches.
Figure 8-7 shows a multihop FCoE topology where the server connects via CNAs to Nexus 5x00s, which in turn connect to Nexus 7000 series switches via the Ethernet carrying FCoE. The storage array is directly connected to the Nexus 7000 via FCoE as well.
Figure 8-7. Multihop FCoE Topology
Storage VDC on Nexus 7000
One of the building blocks in a multihop FCoE topology is the storage Virtual Device Context (VDC) on the Nexus 7000. VDCs are discussed in detail in Chapter 1, “Introduction to Cisco NX-OS,” and the focus in this chapter is on the Storage VDC and its use in a multihop FCoE topology. VDC is a capability of the Nexus 7000 series switches that enables a network administrator to logically virtualize the Nexus 7000 into multiple logical devices. The storage VDC is a special VDC that enables the virtualization of storage resources on the switch. This enables in essence a “virtual MDS” inside the Nexus 7000 that participates fully in the FCoE network as a full fibre channel forwarder (FCF).
With a Storage VDC network, administrators can provide the storage team a context that allows the storage team to manage their own interfaces; configurations; and fibre channel-specific attributes such as zones, zonesets, and aliases. Figure 8-8 shows how a storage VDC can be implanted in an existing topology where single-hop FCoE was initially deployed and then multihop FCoE was added. The storage VDC was created with VE ports connecting downstream to the Nexus 7000 and VE port to the Cisco MDS fibre channel director.
Figure 8-8. Storage VDC on the Nexus 7000
The storage VDC has some requirements that are unique to this type of VDC as storage traffic is traversing it. The first requirement is that the storage VDC can support only interfaces hosted on the F1 or F2/F2e series of modules. These modules support the capability to provide lossless Ethernet and as such are only suitable for doing FCoE. The VDC allocation process in NX-OS does not allow for other types of modules to have interfaces in a VDC that has been defined as a storage VDC.
In addition to requiring F1 or F2/F2e series modules, the storage VDC cannot run nonstorage related protocols. You cannot enable features such as OSPF, vPC, PIM, or other Ethernet/IP protocols in the storage VDC. The only features allowed are directly related to storage. Finally, the default VDC cannot be configured as a storage VDC.