Upgrading and Repairing PCs

SCSI-3

SCSI-3 is a term used to describe a set of standards currently being developed. Simply put, it is the next generation of documents a product conforms to. Unlike SCSI-1 and SCSI-2, SCSI-3 is not one document that covers all the layers and interfaces of SCSI, but is instead a collection of documents that covers the primary commands, specific command sets, and electrical interfaces and protocols. The command sets include hard disk interface commands, commands for tape drives, controller commands for redundant array of inexpensive drives (RAID), and other commands. Additionally, an overall SCSI Architectural Model (SAM) exists for the physical and electrical interfaces, as does a SCSI Parallel Interface (SPI) standard that controls the form of SCSI most commonly used. Each document within the standard is now a separate publication with its own revision level—for example, within SCSI-3 three versions of the SCSI Parallel Interface have been published. Usually, we don't refer to SCSI-3 anymore as a specific interface and instead refer to the specific subsets of SCSI-3, such as SPI-3 (Ultra3 SCSI).

The main additions to SCSI-3 include

• Ultra2 (Fast-40) SCSI

• Ultra3 (Fast-80DT) SCSI

• Ultra4 (Fast-160DT) SCSI

• Ultra5 (Fast-320DT) SCSI

• New Low Voltage Differential signaling

• Elimination of High Voltage Differential signaling

Breaking SCSI-3 into many smaller individual standards has enabled the standard as a whole to develop more quickly. The individual substandards can now be published rather than waiting for the entire standard to be approved.

Figure 8.1 shows the main parts of SCSI today.

The most recent changes or additions to SCSI are the Fast-40 (Ultra2), Fast-80DT (Ultra3), and Fast-160DT (Ultra4) high-speed drives and adapters. These have taken the performance of SCSI up to 320MBps. Also new is the LVD electrical interface standard, which enables greater cable lengths. The older High Voltage Differential signaling has also been removed from the standard.

A number of people are confused over the speed variations in SCSI. Part of the problem is that speeds are quoted as either clock speeds (MHz) or transfer speeds. With 8-bit transfers, you get 1 byte per transfer, so if the clock is 40MHz (Fast-40 or Ultra2 SCSI), the transfer speed is 40MBps. On the other hand, if you are using a Wide (16-bit) interface, the transfer speed doubles to 80MBps, even though the clock speed remains at 40MHz. With Fast-80DT, the bus speed technically remains at 40MHz; however, two transfers are made per cycle, resulting in a throughput speed of 160MBps. The same is true for Ultra4 SCSI, which runs at 80MHz, transfers 2 bytes at a time, and has two transfers per cycle. Ultra4 is also called Ultra320 and is the fastest form of parallel SCSI available today—the Ultra5 (Ultra640) standard is still under development.

Finally, confusion exists because SCSI speeds or modes are often discussed using either the official terms—such as Fast-10, Fast-20, Fast-40, and Fast-80DT—or the equivalent marketing terms, such as Fast, Ultra, Ultra2, and Ultra3 (also called Ultra160). Refer to Table 8.2 for a complete breakdown of SCSI official terms, marketing terms, and speeds.

The further evolution of the most commonly used form of SCSI is defined under the SPI standards within SCSI-3. The SPI standards are detailed in the following sections.

SPI or Ultra SCSI

The SCSI Parallel Interface standard was the first SCSI standard that fell under the SCSI-3 designation and is officially known as ANSI X3.253-1995. SPI is also called Ultra SCSI by most marketing departments and defines the parallel bus electrical connections and signals. A separate document called the SCSI Interlock Protocol (SIP) defines the parallel command set. SIP was included in the later SPI-2 and SPI-3 revisions and is no longer carried as a separate document. The main features added in SPI or Ultra SCSI are

• Fast-20 (Ultra) speeds (20MBps or 40MBps)

• 68-pin P-cable and connectors defined for Wide SCSI

SPI initially included speeds up to Fast SCSI (10MHz), which enabled transfer speeds up to 20MBps using a 16-bit wide bus. Later, Fast-20 (20MHz), commonly known as Ultra SCSI, was added through an addendum document (ANSI X3.277-1996), allowing a throughput of 40MBps on a 16-bit wide bus (commonly called Ultra/Wide).

SPI-2 or Ultra2 SCSI

SPI-2 is also called Ultra2 SCSI, was officially published as ANSI X3.302-1998, and adds several features to the prior versions:

• Fast-40 (Ultra2) speeds (40MBps or 80MBps)

• Low Voltage Differential signaling

• Single Connector Attachment (SCA-2) connectors

• 68-pin Very High Density Connector (VHDC)

The most notable of these is a higher speed called Fast-40, which is commonly called Ultra2 SCSI and runs at 40MHz. On a narrow (8-bit) bus, this results in 40MBps throughput, whereas on a wide bus (16-bit), this results in 80MBps throughput and is commonly referred to as Ultra2/Wide.

To achieve these speeds, a new electrical interface called LVD must be used. The slower single-ended electrical interface is only good for speeds up to Fast-20; Fast-40 mode requires LVD operation. The LVD signaling also enables longer cable lengths up to 12 meters with multiple devices or 25 meters with only one device. LVD and SE devices can share the same cable, but in that case, the bus runs in SE mode and is restricted in length to as little as 1.5 meters in Fast-20 mode. LVD operation requires special LVD-only or LVD/SE multimode terminators. If multimode terminators are used, the same terminators work on either SE or LVD buses.

The SPI-2 standard also includes SCSI Interlink Protocol (SIP) and defines the Single Connector Attachment (SCA-2) 80-pin connector for hot-swappable drive arrays. There is also a new 68-pin Very High Density Connector (VHDC), which is smaller than the previous types.

SCSI Signaling

"Normal," or standard, SCSI uses a signaling technique called single-ended (SE) signaling. SE signaling is a low-cost technique, but it also has performance and noise problems.

Single-ended signaling is also called unbalanced signaling. Each signal is carried on a pair of wires, usually twisted to help reduce noise. With SE, one of the pair is grounded—often to a common ground for all signals—and the other carries the actual voltage transitions. It is up to a receiver at the other end of the cable to detect the voltage transitions, which are really just changes in voltage.

Unfortunately, this type of unbalanced signaling is very prone to problems with noise, electromagnetic interference, and ground leakage; these problems get worse the longer the cable is. This is why Ultra SCSI was limited to such short maximum bus lengths—as little as 1.5 meters, or 5 feet.

When SCSI was first developed, a signaling technique called High Voltage Differential signaling was also introduced into the standard. Differential signaling, also known as balanced signaling, is still done with a pair of wires. In fact, the first in the pair carries the same type of signal that single-ended SCSI carries. The second in the pair, however, carries the logical inversion of that signal. The receiving device detects the difference between the pair (hence the name differential). By using the wires in a balanced pair, the receiver no longer needs to detect voltage magnitude, only the differential between voltage in two wires. This is much easier for circuits to do reliably, which makes them less susceptible to noise and enables greater cable length. Because of this, differential SCSI can be used with cable lengths of up to 25 meters, whereas single-ended SCSI is good only for 6 meters maximum, or as little as 1.5 meters in the faster modes.

Figure 8.2 shows the circuit differences between balanced (differential) and unbalanced (single-ended) transmission lines.

Unfortunately, the original standard for HVD signaling called for high-voltage differentials between the two wires. This meant that small, low-power, single-chip interfaces using HVD signaling could not be developed. Instead, circuits using several chips were required. This worked at both ends, meaning both the host adapter and device circuitry had to be larger and more expensive.

Another problem with HVD SCSI is that although the cables and connectors look (and are) exactly the same as for SE SCSI, both types of devices can't be mixed on the same bus. If they are, the high voltage from the HVD device burns out the receiver circuits on all SE devices attached to the bus. In other words, the result is smoked hardware—not a pretty sight.

Because SE SCSI worked well enough for the speeds that were necessary up until recently, HVD SCSI signaling never really caught on. It was used only in minicomputers and very rarely, if at all, in PCs. Because of this fact, the extra cost of this interface, and the fact that it is electrically incompatible with standard SE SCSI devices, HVD signaling was removed from the SCSI specification in the latest SCSI-3 documents. So, as far as we are concerned, it is obsolete.

Still, a need existed for a more reliable signaling technique that would allow for longer cable lengths. The answer came in the form of LVD signaling. By designing a new version of the differential interface, it can be made to work with inexpensive and low-power SCSI chips. Another advantage of LVD is that because it uses low voltage, if you plug an LVD device into an SE SCSI bus, nothing will be damaged. In fact, as an optional part of the LVD standard, the LVD device can be designed as a multimode device, which means it works on both SE and LVD buses. In the case of installing a multimode LVD device into an SE bus, the device detects that it is installed in an SE bus and defaults to SE mode.

Therefore, all multimode LVD/SE SCSI devices can be used on either LVD or SE SCSI buses. However, when on a bus with even one other SE device, all the LVD devices on the bus run only in SE mode. Because SE mode supports only SCSI speeds of up to 20MHz (Fast-20 or UltraSCSI) and cable lengths of up to 1.5 or 3 meters, the devices also work only at that speed or lower; you also might have problems with longer cables. Although you can purchase an Ultra3 SCSI multimode LVD/SE drive and install it on a SCSI bus along with single-ended devices, you will certainly be wasting the capabilities of the faster device.

Note that all Ultra2 and Ultra3 devices support LVD signaling because that is the only way they can be run at the Ultra2 (40MHz) or Ultra3 (80MHz) speeds. Ultra SCSI (20MHz) or slower devices can support LVD signaling, but in most cases, LVD is synonymous with Ultra2 or Ultra3 only.

Table 8.2, shown earlier, lists all the SCSI speeds and maximum lengths for each speed using the supported signaling techniques for that speed.

Because the connectors are the same for SE, HVD, LVD, or multimode SE/LVD devices, and because putting an HVD device on any bus with SE or LVD devices causes damage, it would be nice to be able to tell them apart. One way is to look for a special symbol on the unit; the industry has adopted different universal symbols for single-ended and differential SCSI. Figure 8.3 shows these symbols.

If you do not see such symbols, you can tell whether you have a High Voltage Differential device by using an ohmmeter to check the resistance between pins 21 and 22 on the device connector:

• On a single-ended or Low Voltage Differential device, the pins should be tied together and also tied to the ground.

• On a High Voltage Differential device, the pins should be open or have significant resistance between them.

Although you will blow up stuff if you plug HVD devices into LVD or SE buses, this generally should not be a problem because virtually all devices used in the PC environment are SE, LVD, or LVD/SE. HVD has essentially been rendered obsolete because it has been removed from the SCSI standard with Ultra3 SCSI (SPI-3).

SPI-3 or Ultra3 SCSI (Ultra160)

SPI-3—also known as Ultra3 or Ultra160 SCSI—builds on the previous standard and doubles the speed again to Fast-80DT (double transition). This results in a maximum throughput of 160MBps. The main features added to SPI-3 (Ultra3) are

• DT (double transition) clocking

• Cyclic redundancy check (CRC)

• Domain validation

• Packetization

• Quick Arbitrate and Select (QAS)

Double transition clocking sends data on both the rising and falling edges of the REQ/ACK clock. This enables Ultra3 SCSI to transfer data at 160MBps, while still running at a bus clock rate of 40MHz. This mode is defined for 16-bit wide bus use only.

Cyclic redundancy checking (CRC) is a form of error checking incorporated into Ultra3 SCSI. Previous versions of SCSI used simple parity checking to detect transmission errors. CRC is a much more robust form of error-detection capability that is far superior for operation at higher speeds.

Domain validation allows better negotiation of SCSI transfer speeds and modes. With prior SCSI versions, when the bus is initialized, the host adapter sends an INQUIRY command at the lowest 5MHz speed to each device to determine which data-transfer rate the device can use. The problem is that, even though both the host adapter and device might support a given speed, there is no guarantee that the interconnection between the devices will reliably work at that speed. If a problem occurs, the device becomes inaccessible. With domain validation, after a maximum transfer speed is negotiated between the host and device, it is then tested at that rate. If errors are detected, the rate is stepped down until the connection tests error-free. This is similar to how modems negotiate transmission speeds before communicating and will go a long way toward improving the flexibility and perceived reliability of SCSI.

Packetization is a protocol that enables information to be transferred between SCSI devices in a much more efficient manner. Traditional parallel SCSI uses multiple bus phases to communicate different types of information between SCSI devices: one for command information, two for messages, one for status, and two for data. In contrast, packetized SCSI communicates all this information by using only two phases: one for each direction. This dramatically reduces the command and protocol overhead, especially as higher and higher speeds are used.

Packetized SCSI is fully compatible with traditional parallel SCSI, which means packetized SCSI devices can reside on the same bus as traditional SCSI devices. As long as the host adapter supports the packetization, it can communicate with one device using packets and another using traditional protocol. Not all Ultra3 or Ultra160 SCSI devices include packetization support, however. Ultra3 devices that support packetization typically are referred to as Ultra160+ SCSI.

Quick Arbitrate and Select (QAS) is a feature in Ultra3 SCSI that reduces arbitration time by eliminating bus free time. QAS enables a device to transfer control of the bus to another device without an intervening BUS FREE phase. SCSI devices that support QAS report that capability in the INQUIRY command.

Ultra160 and Ultra160+

Because the five main new features of Ultra3 SCSI are optional, drives could claim Ultra3 capability and not have a consistent level of functionality. To ensure truth in advertising and a minimum level of performance, a group of manufacturers got together and created a substandard within Ultra3 SCSI that requires a minimum set of features. These are called Ultra160 and Ultra160+ because both indicate 160MBps throughput. These new substandards are not an official part of the SPI standard. Even so, they do guarantee that certain specifications will be met and certain performance levels will be attained.

Ultra160 is a specific implementation of Ultra3 (SPI-3) SCSI that includes the first three additional features of Ultra3 SCSI:

• Fast-80DT clocking for 160MBps operation

• CRC

• Domain validation

Ultra160 SCSI runs in LVD mode and is backward compatible with all Ultra2 SCSI (LVD) devices. The only caveat is that no SE devices must be on the bus. When Ultra2 and Ultra160 (Ultra3) devices are mixed, each device can operate at its full-rated speed independent of the other. The bus will dynamically switch from single- to double-transition mode to support the differences in speeds.

Ultra160+ adds the other two features, ensuring a full implementation of Ultra3:

• Packetization

• Quick Arbitrate and Select

With Ultra160 and Ultra160+, a known level of functionality ensures that a minimum level of performance will be met. Ultra160+ SCSI is the highest-performance PC-level storage interface and is best suited for high-traffic environments, such as high-end network servers or workstations. The adaptability and scalability of the interface enables high performance with high reliability.

SPI-4 or Ultra4 SCSI (Ultra320)

SPI-4, also known as Ultra4 or Ultra320 SCSI, has all the same features as the previous Ultra3 (Ultra160) and adds several new features to ensure reliable data transmission at twice the speed.

Ultra320 SCSI integrates both the Packetization and Quick Arbitrate and Select features from Ultra160+ SCSI as mandatory features.

Ultra320 SCSI then adds the following new features:

• Transfer speed. Ultra320 transfers data 2 bytes (16 bits) at a time at 80MHz using double-transition (DT) cycling, meaning it transfers twice per cycle (hertz). This results in a burst transfer rate of 320MBps.

• Read/Write data streaming. This minimizes the overhead for queued data transfers by enabling a device to send one data stream queue-tag packet followed by multiple data packets. Previously, only one data packet could be sent with each queue-tag packet. Write performance is also increased because there are fewer bus turnarounds from data in to data out.

• Flow control. This allows a target device to indicate when the last packet of a data stream will be transferred, which enables the initiator to terminate the data prefetch or begin flushing data buffers sooner than previously possible.

SPI-5 or Ultra5 SCSI (Ultra640)

Work has begun on the SPI-5 SCSI standard, also called Ultra5 or Ultra640 SCSI. All that is known at this time is that it is based on the Ultra320 standard but will transfer at twice the speed—an amazing 640MBps.

RAID Arrays

Most servers, especially at levels above the workgroup, use SCSI drives rather than ATA drives because of their superior performance. You can enhance performance and data reliability further by creating a drive array. RAID (redundant array of individual drives) technologies are used by both SCSI and ATA drives. Current SCSI-based RAID products primarily support Ultra320 and Ultra160 drives and are used in traditional servers and rackmounted computers. For a complete description of RAID levels and terminologies, see Chapter 7, "The ATA/IDE Interface."

Fibre Channel SCSI

Fibre Channel SCSI is a specification for a serial interface using a fibre channel physical and protocol characteristic, with a SCSI command set. It can achieve 200MBps or 400MBps over either fiber or coaxial cable of several kilometers in length. Fibre Channel is designed for long-distance connectivity (such as several kilometers) and connecting multiple systems, and it has become a popular choice for storage area networks (SANs) and server clusters. Fibre Channel SCSI complements, rather than replaces, Ultra160 and Ultra320 SCSI, which are designed for direct connection to servers.

200MBps versions of Fibre Channel SCSI use the gigabit interface connector (GBIC), whereas 400MBps versions use either the small form factor pluggable (SFP) connector for optical connections or the high-speed serial data connector (HSSDC) for copper cable.

iSCSI

The latest variation on SCSI, iSCSI, combines the performance of SCSI drives with Ethernet networking up to gigabit speeds. Because iSCSI uses Ethernet to transport data between systems, iSCSI storage can be located anywhere an Ethernet network can reach, including Internet access. In addition, iSCSI storage enables secure remote storage for computers that could be hundreds of kilometers away. Because iSCSI data can be routed the same way any other type of Ethernet data can be routed, it enables data to be transported even when some connections between the server and the storage devices are unavailable.

Eventually, iSCSI is expected to replace Fibre Channel in uses such as network attached storage (NAS), SANs, and storage clusters. Similar to Fibre Channel, iSCSI cards can be purchased in both copper-wire and fiber-optic versions to match the Ethernet network already in use. The first iSCSI devices were produced by Adaptec and Cisco Systems in mid-2002.