Network virtualization is now a hot topic, and major manufacturers are scrambling to grab this cake. Technologies involved include virtual Emulation for NICs, I/O pass-through, SR-IOV), virtual access technology for networks VN-Tag/VEPA), covering network switching VXLAN/NVGRE ), and Software Defined network SDN/OpenFlow. For details, refer to the blog "Network virtualization-ongoing network revolution" published by EMC China Research Institute, which summarizes the status quo of network virtualization.
I don't know if you have noticed that the network mentioned here is actually a narrow data communication network, including LAN and WAN/Internet ), the protocols used include Ethernet, Layer 2), Internet IP, Layer 3), and so on. However, there is also an important network in enterprise-level data centers or cloud computing data centers, that is, the storage network StorageNetwork). The dominant network is the storage Domain Network SAN Based on FC technology ). The Virtualization of storage networks seems to be a neglected field. Although some work has been carried out in this area, there is still not much attention in general. What is the prospect of Storage Network virtualization? Is it the Great Northern land that has never been reclamation )?
Storage Network Overview
A storage Network, also known as the storage Domain Network StorageArea Network (SAN), is a dedicated Network that provides unified access to block-level data storage. The supporting technology is fiber channel (FC. FC is not the only choice for building a storage network, but a dominant technology in the industry. FC usually includes two meanings: 1) transmission media, which can provide long-distance, high-speed, low-latency, low-error rate transmission, and hardware HBA) for transmission protocol note: here the optical fiber-optic) is also not a must); 2) the transmission Protocol fiber Channel Protocol, FCP), a layer-5 network Protocol stack FC-0 to the FC-4, we can look at the seven-layer protocol stack of ISO ). About FCP here only mention the more important two layers: 1) The FC-2 is the data transmission layer, where the data frame traffic control flowcontrol); 2) the FC-4 is the upper layer protocol ing layer, maps application layer protocols such as SCSI and IP addresses. Careful readers may find problems. Why is there an IP protocol? Here is an episode. The FC protocol was originally designed as the backbone network backbone for LAN) to replace the Mbps Ethernet ). Of course, we all know that the Gigabit Ethernet and 10-Gigabit Ethernet technologies are the best.
The LAN Based on Ethernet technology is familiar to everyone. So what are the similarities and differences between the SAN Based on Fiber Channel Technology and the SAN Based on Ethernet technology? Figure 1 simplifies the LAN and SAN in the data center. As you can see, Switching is also used in SAN construction ). In fact, most of the concepts between the two are analogous. Here we only mention two key differences. First, the LAN is a fully connected network, and communication may exist between any two node hosts. SAN is a binary graph Group 1: Host, group 2: from the graph theory perspective: storage System. Second, LAN is a network that best serves best-effort. Congestion and packet loss may occur. Reliable Data Transmission relies on upper-layer protocols such as TCP ); SAN provides very reliable data transmission, and the upper-layer protocol SCSI) is much less sensitive to packet loss.
Note: a bipartite graph refers to a graph where a vertex can be divided into two non-intersecting sets so that the vertex in the same set is not adjacent.
Figure 1 LAN) and Storage Network SAN)
Table 1 LAN) and Storage Network SAN)
The basic communication of FC mainly exists between two types of ports, namely, N-PortNodePort) and F-PortFabricPort ). F-Port is the Port on the FC switch and is used to connect to N-Port. N-Port is the Port of the host or storage system of the terminal device. It is usually located on the host Adapter HostBus, HBA, figure 2) for access to the switching network. Each N-Port has a 64-bit globally unique Name WorldwidePort Name, WWPN, refer to the MAC address of the Ethernet Card ). After the N-Port is connected to the network, a 24-bit Port number (N-PortID) is assigned, which is similar to the IP address obtained by DHCP ). This N-PortID is used for addressing and shipping of all subsequent data frames.
Optical fiber memory card fc hba) Virtualization
The first step of network virtualization is to connect devices to a virtual machine through multi-tenant multi-tenancy. Users who have some knowledge about Data Communication Network virtualization or server virtualization should be impressed with the virtual network card in the virtual machine. So should we also provide a virtual memory card for virtual machines in the Process of Storage Network virtualization? The answer is double, and yes or no.
Figure 2 Optical Fiber adapter fc hba)
The answer may not seem so intuitive, but it is a realistic choice so far. The underlying reason is that applications use different types of networks. We know that applications will directly deal with layer-2 Ethernet or layer-3 IP addresses when using communication networks. When using a storage network, the difference is that there is a layer of SCSI protocol in the middle, and most applications perform read and write operations) is actually the SCSI disk at least looks like ), the underlying FC network protocols are completely hidden by virtual systems such as VMwarevSphere. This is why Nic virtualization is necessary, and HBA virtualization needs to be less direct. Without a virtual HBA, how does a VM directly access the LUN in the storage system? We will answer this question in the next section about NPIV.
In this case, is there a solution that supports the virtual HBA so far )? As a matter of fact, NPIV has its inherent shortcomings, which only satisfies basic I/O requirements and cannot meet the needs of all applications. The discussion about the virtual HBA will be completed after the introduction of NPIV.