Storage Area Networks

In Figure 11.1 you saw that a SAN operates as a file server on a separate network from the LAN. A SAN also has some advantages and benefits, based on its capabilities, when compared with NAS:

• SANs generally provide a faster response to a server than even some versions of locally connected SCSI devices. One version or another of SCSI technology is used in most standard servers/workstations today. SCSI has continued to be upgraded with newer versions, but the storage capacity that can be provided has reached its maximum in many environments.

• SANs can use bridges to existing SCSI equipment, thus preserving your existing investment in expensive devices. You don't have to throw out those older SCSI devices (or those you purchased yesterday...). Instead, a bridge can allow you to continue to use this hardware as long as it remains cost-effective.

• It would be difficult to limit the amount of storage a SAN can hold. Both Arbitrated Loops and Fabric Switched technologies (both of which are described later in this chapter) have their own methods of operation. Each of these solutions, or even a mixture of the two, can provide enough storage to meet nearly any computing environment today. Storage is practically unlimited for a SAN.

• Although it is possible to use copper cabling with many SANs, it is more common to use fiberoptic cabling. The term Fibre Channel was developed to separate the protocol from the fiberoptic cable; yet today it is associated more with fiber-optic cables than with copper cables.

• A very important feature of SANs is the capability of eliminating the backup window discussed earlier. NAS devices have their own CPU just as do SAN-enabled adapter cards. Yet if you have very large amounts of critical data that you want to back up, a SAN will definitely provide a better solution. SAN devices, in general, operate better at creating backups that don't impact the network. This is especially true because the SAN device can do the backup while continuing to offer file services to clients or servers, just as NAS does. However, the SAN Fibre Channel (or IP SAN network) can satisfy the server's requests faster than local SCSI attached storage, often allowing a SAN to perform backup operations with very little or no noticeable impact on the clients.

• The Fibre Channel protocol and associated hardware devices are the main choice for SANs implementers today. However, a newer technology, called IP SANs, promises to offer about the same performance as a Fibre Channelbased SAN. IP SANs will also offer minimal training (as opposed to Fibre Channel SANs) because employees will be using existing or similar devices such as IP switches and routers. One method of implementing an IP SAN is the use of iSCSI, which uses a TCP/IP network to transmit data between a iSCSI-enabled SAN device and a server.

As you can see, SANs are fundamentally different from NAS storage. The main advantages that SANs based on Fibre Channel enjoy is the minimal overhead involved in Fibre Channel, the low-level error detection and correction techniques, and the fast fiber-optic transmission rates on a separate network. Fibre Channel protocols currently run at speeds up to 2Gps. On the other hand, IP SANs are less expensive to implement than Fibre Channel because they can use existing copper or fiber-optic Ethernet networks. IP SANs use their own block-oriented host bus adapters to transfer data more quickly than standard network adapters do.

SAN and NASMix and Match

It is possible to use both NAS and SANs on the same network. For some departmental servers, where traffic is limited to the local LAN segment, using a file server with a NAS device for additional storage is an excellent choice, especially from an expense viewpoint. If your company has a large Internet presence, you may want to consider using a SAN to back up that important presence.

Most large networks consist of small departments as well as departments or applications that require huge amounts of mission-critical storage on demand. In this case a mix of NAS devices (for local departments) and SANs storage (for mission-critical storage support) may be your best solution.

The point to make is that NAS and SANs are not exclusive. Each has multiple functions to fulfill, and each performs using different techniques. The good news is that you can use both on your network at the same time.

Using Fibre Channel as a Network Transport

Most Storage Area Networks today use Fibre Channel to connect to a wide variety of SAN products. Fibre Channel is generally used across fiber-optic cables, although the specifications do allow for a short-distance copper-cable haul.

Fibre Channel uses a serial form of communications. Between any two devices, there are two connections: one to transmit data and one to receive data. The two cables are swapped so that the transmitter of one device is connected to the receiver of the other end of the connection.

With the use of two cables, it is also possible for information to be flowing in two directions at the same timefull-duplex communications. Many Fibre Channel functions are performed on the adapter itself (called a host bus adapter, although there are others), instead of those functions that a typical IP NIC passes up to software-based code to interpret and act on.

As if fiber-optic cabling were not fast enough, Fibre Channel also offers an encoding technique that has been around for years, used in other communication techniques used on both copper and fiberoptic cabling. This encoding is referred to as 8B/10B encoding. It was not selected by random from among the many choices. Instead it was selected because of its unique properties that provide a lot of error checking and long-distance serial communications.

Encoding Data on Fibre Channel Networks

Most network professionals think of a typical network packet, or datagram or frame, whatever level you are working at, to be a simple collection of bytes. However, if you are going to step into Fibre Channel, be aware that you won't find that convention used here. Instead, 8-bit characters are translated to 10-bit transmission characters. Although it may seem odd to convert 8 bits to 10 bits to achieve a high transmission rate, there are several good reasons for using this technique, discussed later in this chapter.

The main feature that distinguishes 8B/10B encoding from other standard LAN protocols is that it attempts to keep the number of zeros and ones transmitted on the network media to an equal number. This is called neutral disparity. Each transmission character to be used is chosen based on the current disparity (running disparity) of the serial communications. Only some of the transmission characters contain an equal number of zeros and ones. Others contain more zeros than ones, or vice versa. If a transmission character contains more ones than zeros, it is said to have positive disparity. If the transmission character has more zeros than ones, it is said to have negative disparity.

So if the last transmission character sent out onto the network media had a negative disparity, then the adapter card circuitry would choose a transmission character with positive disparity to keep the running disparity on the line neutral over a short time. Some 8-bit values translate into a transmission character with neutral disparity, whereas others translate into two characters: one with positive disparity and another with negative disparity. Those that translate to two values allow the adapter card to choose a character to help maintain neutral disparity.

What begins as a simple byte undergoes a transformation before being sent out on a Fibre Channel transmission. Fibre Channel uses a serial mechanism for sending bits out onto the networkone bit at a time, compared to a parallel environment (such as SCSI) in which several wires are used, each for a single bit, and a timing wire is used to inform the other end of the data transfer that a byte (or more) has been received.

Because serial data transfers occur only in one direction on a single cable (which is why you have send/receive cables on each host bus adapter), another method must be used to recover the clock. This is done by using special code, called a comma character, and by limiting the number of bits that can be transmitted in a row. The recipient of the data transfer can recover quickly if the correct number of bytes for a time period is not transmitted correctly, or if the comma character is sent through the communications channel.

The technique of 8B/10B encoding is beyond the scope of this book. However, as discussed previously, these are the important things about 8B/10B encoding:

• No more than four of the same bit (0 or 1) are ever sent in a row (other than for the comma character).

• Each 8-bit quantity can translate to either one or two 10-bit transmission characters. The character used depends on the disparity of the network.

• Disparity means that an equal number of zeros and ones are being transmitted within a short period. 8B/10B encoding chooses transmission characters that attempt to maintain neutral disparity, or an equal number of zeros and ones.

As you can see, the concept of disparity is crucial to the Fibre Channel protocol. There are several reasons for this. One reason is that lasers and fiber-optic transmitters can overheat. If the laser can be used only about 50% of the time, this can make equipment less expensive to produce. Thus, alternating between zeros and ones on a periodic basis can help keep heat to a minimum.

Another reason for guaranteeing a state change from more zeros to ones, and for using the special comma character, is to enable the receiver to recover the clock. Because there is no extra wire or signal to synch the timing between the two network adapters, another method must be used on the receiving end so that it can determine when it has received a transmission character. By forcing a change on the line within a guaranteed length of time, it is easy for the receiver to pick up after a garbled character and continue receiving data.

Pushing the Protocol Stack Down to the Adapter Level

The Host Bus Adapter (HBA) is the terminology given to the adapter card that connects a computer to a Fibre Channel SAN. The HBA differs from an Ethernet card in that the HBA performs more functions than the Ethernet card, which frees up CPU cycles for other duties. The Fibre Channel protocol elements are processed on the HBA. Fibre Channel can accept many different upper-level protocols, such as IP, SCSI, and HIPPI, and can transfer their frames as a payload of the Fibre Channel frame.

The Fibre Channel frame format also has a larger payload than a standard IP frame, as well as a lower header-to-payload ratio. In Fibre Channel the final frame is put together by the HBA and transmitted on the network media.

Much error recovery is done at the physical level, but the simplicity of the Fibre Channel protocols is such that errors occur at a very low rate. The running neutral disparity allows for a quick recovery in case of problems on the line.

Fibre Channel Over IP

Fibre Channel can cover distances measured in kilometers. When it is necessary to cover a larger distance, Fibre Channel frames can be encapsulated inside an IP packet and transmitted through an ordinary IP network. This does not imply a translation of the Fibre Channel frame to an IP frame. Instead, the Fibre Channel frame is stored in the payload of the IP frame, in the same manner that other protocol frames are carried as the payload of the Fibre Channel frame. This makes it possible to access data over long distances. It also enables you to create SANs that mirror each other in separate data centers for disaster recovery purposes.