Inheriting the architecture of Fibre Channel, SAN also has three topologies: exclusive point-to-point connection, providing guaranteed bandwidth; arbitration circuit, providing two or more device sharing bandwidth; Switch Fabric provides simultaneous performance. The dedicated point-to-point connection point-to-point connection is the most basic, the simplest architecture, two N_Ports direct docking, one N_PORT transmitting end (TX) receives another N_Port receiving end (RX), in turn The receiving end is connected to the transfer end of others. This architecture basically can only establish a system with only two devices. Of course, these two devices also have all bandwidth (100MB / s, or 200MB / s two-way transmission). Note: n_port: The device of each Fiber Channel is called Node, each Node has a (or more) connection (port (s)), each connection in Node is called n_port, that is, Node Port. In terms of practical applications, the chances of system can do full duplex transmission are not very large. For example, the server is on the same time, not from the storage system, is to write data into the storage system; it is too little to write the opportunity to write while reading at the same time. Moreover, the bandwidth of its PCI BUS is not 132Mb / s, which is not practical. Therefore, the point-to-point connection architecture, although it is also one of SAN, but in order to directly connect to the storage system (Direct-Atached Storage), only the SCSI or IDE is replaced with FIBRE CHANNEL. Since the technology of Arbitrated Loop and Switch Fabric is increasing, its product price is also increasingly economical, so the SAN system is basically made from Arbitrated Loop architecture to the Switch Fabric architecture. The architecture of point-to-point connection is only suitable for the initial setting of the storage system. When the capacity needs are not very large, be a conservative investment, but retain the expansion capacity of the future system (Scalability). But when choosing the host adapter card (HBA: Host Bus Adapter) and the surrounding, you must pay special attention to its specifications, determine the drivers and firmware (FIRMWARE, Microcode) that can support the future to Arbitrated Loop and Switch Fabric Ability to avoid waste of investment, or even the storage system must reset the pain. Arbitrated LOOP Arbitration Retraction Arbitration Project, the specific point-to-point connection architecture has more flexibility, one loop can connect to 127 devices; at the same time, its unit cost is low, so the LOOP architecture is currently the most A widely used architecture.
The Arbitrated LOOP entity architecture is like existing FDDI (fiber distributed data interface) and Token Ring, the first NL_PORT transmitted to the second NL_PORT receiver; second NL_Port The transfer end is connected to the receiving end of the next NL_Port; in this type, the transfer end of the last NL_Port is connected to the receiving end of the NL_PORT. Thus, a closed loop is formed. Note: NL_PORT: A n_port If it is connected to Arbitrated Loop, it is called NL_Port, which is the meaning of node loop port. All devices in an Arbitrated Loop share the overall bandwidth (100MB / s). Therefore, the larger the circuit architecture, the more the total number of devices), and the bandwidth to each device is relatively small. Of course, at the same time point, it is not a transmission data on each device on the loop. Therefore, the number of active nodes determines the average bandwidth of the entire loop. For example, in two servers connect 40 JBOD disk drives (Just A Bunch of Disks, a bunch of independent disk drives, not built into RAID), if each server's HBA can perform 8 Concurrent Threads (Concurring thread), there may be 16 active nodes at the same time, so the bandwidth of each node can share is about 6.6Mb / s, which is quite grievable for performance of more than 30MB / s for the current Fibre Channel disk drive. of.
Early Arbitrated Loop is real with single-line Cable (fiber or copper), a Node Node, which is the way, the so-called daisy chain, forms a closed loop. This approach is the most direct, most economical, but also the highest risk, the most unreliable architecture. Because if there is any Cable break, or any node is faulty or lost the power, the entire loop is completely. Some SAN's devices, with the Bypass line, when you fail or shut down it, you can jump over the loop, so that the system reliability can be improved, but it is still unable to solve the problem of CABLE. In order to solve the above problems, there is an advent of Arbitrate Loop Hub. Loop hub basically is built into the body line, the hardware line connects the first NL_PORT TX RX, so that it is connected, and finally connects the last NL_PORT Tx to the first NL_Port RX. . Loop Hub's best feature is that each port has a BYPASS, which automatically occurs a BYPASS role when connecting or turning off when connecting to any NL_PORT. In addition, the basic functions of the LOOP HUB include: the ability to detect the addition and removal of the device, the correct line is added or removed, and supports dynamic reintegraph.
With the Arbitrated Loop established by Loop Hub, basically form a star-shaped loop architecture that solves the above reliability and provides more elasticity and convenience: the wiring work is more convenient when the system is built; When the device is faulty, you can directly remove the maintenance, do not need to turn off the entire loop; you can add the device, even a new loop hub (cascading) to expand the system. For example, connect a JBOD disk cabinet to a NL_Port of Loop Hub, or a new loop HUB, which is Cascade, and adds a number of disk drives to the loop. Arbitrated loop hub's cascading is very convenient, no special connection line, just pick a NL_Port to pick up a NL_Port as long as the general FIBRE CHANNEL CABLE can be used. Do not use Crossover Cable or specific ports to do Cascading using Crossover Cable or specific ports. San Switch Fabric Switch Fabric is an interleaving network system that uses Fibre Channel Switch as the main trunk. Every port in Switch has independent bandwidth, the most important feature of Fabric, is to make multiple transfers simultaneously. The effective bandwidth of the entire Fabric is the sum of the links that can be established simultaneously. For example, a 16-port Switch can support up to 8 simultaneous online, with a total bandwidth of 800MB / s (or 1.6Gb / s Full Duplex). In order to effectively enhance the performance of Fabric Switch, Fibre Channel Switch is generally applying the Cut-Through exchange mechanism. Cut-Through technology, when the information is sent into Switch, first discriminate its destination address, and immediately transferred to the destination, rather than waiting until the entire information is received, and then discriminating (Store) -AND-forward. As such, the FRAME can be forwarded in the shortest delay time.
