Ultra SCSI White Paper Introduction This paper discusses a new feature called Ultra SCSI. SCSI (Small Computer System Interface) is the best known and most widely used ANSI (American National Standards Institute) interface. The purpose of this paper is to show how SCSI has evolved over the years, and inspite of some limitations existing today, parallel SCSI has many advantages over the other leading interfaces. Several alternatives are discussed along with the unique advantage of Ultra SCSI which continues to evolve. Ultra SCSI and Other Competitive Interfaces The goal of any peripheral interface is to provide reliable connectivity and avoid being a bottleneck. Just as the internet has seen dramatic increases in traffic, the systems which serve the internet are being called on to deliver unprecedented volumes of data. Internet applications, and others such as video-on-demand, impact the drive in two ways. First, the drive itself must be faster, which in turn puts strain on the interface to keep up. Second, the sheer volume of data being stored in some systems requires a much larger number of drives than ever before. Some video-on-demand sites now being set up call for over 1,500 drives in a single subsystem. These drives are all being connected with the SCSI interface using a large number of SCSI buses. Clearly this interface has come a long way, but the market is demanding a giant step forward. Also, since new generations of high-speed microprocessors are emerging and with disk density increasing, hard disk drive manufacturers are rolling out drives with faster read channels and higher sustained data transfer rates. Improving internal data rates, however, creates a need for faster external (buffer-to-host) data transfer rates, since using even two of these super fast drives can saturate a bus, such as the SCSI-2 narrow bus whose bandwidth permits data transfers of only 10 MB/sec. These are the main reasons for the feverish activity to develop a serial SCSI. A technology capable of boosting hard drive data transfer rates into the 100 MB/sec range eliminating the spectre of interfaces that may otherwise create a system performance bottleneck. However, the transition to a new serial drive interface won't be fast, smooth, or inexpensive. The serial SCSI approaches currently in contention are - FC-AL (Fibre Channel-Arbitrated Loop) which is a subset of the Fibre Channel network systems interconnection standard adopted by the ANSI, and SSA (Serial Storage Architecture). Since adopting any of these interfaces translates into considerable work and costs for systems manufacturers, many will elect to postpone their decision rather than lock themselves into a potentially obsolete technology. And therein lies the catch: the System OEMs need a high I/O throughput solution now, not two or three years from now. Fortunately such a solution has arrived in the form of the new Ultra SCSI interface. An extension of SCSI-2, first proposed by the ANSI XT310 committee as Fast-20. Ultra SCSI represents an enhancement of the current parallel SCSI interface standard that doubles SCSI drives' burst data transfer rates while maintaining backward compatibility. A Brief History about the Small Computer System Interface SCSI was created to satisfy the need for a more flexible, faster, command-controlled interface for hard disk drives and other computer peripherals. Despite the term "small" in its name, SCSI is large. It's large in use, in market impact, influence, and unfortunately, in documentation. Early SCSI was simple, it had a few dozen commands, and the specification could be tucked in your pocket. The standards effort that began with a 20-page specification in 1980 has grown to a 600-page extravaganza of technical information. In 1985, a group of manufacturers approached the X3T9.2 Task Group when the first SCSI standard was being finalized as an ANSI standard. The group wanted to increase the mandatory requirements of SCSI and define further features for direct-access devices. Rather than delay the SCSI standard, the Task Group formed an ad hoc group to develop a working paper that was eventually called the Common Command Set (CCS). Many products were designed to this working paper. The standard provided what were at the time, marvelous new performance characteristics- it could connect a host with as many as 7 different devices, and data could be moved at an exhilarating 5 MB/sec. Even before ANSI published the first SCSI standards document in 1986, the ASC (Accredited Standards Committee) X3T9.2 was hard at work on SCSI-2. While SCSI-2 was to go beyond the original SCSI standard (now referred to as SCSI-1), it was to retain a high degree of compatibility with SCSI-1 devices. No technical rationale can be offered as to where SCSI-1 ended and SCSI-2 began, or as to where SCSI-2 ended and SCSI-3 began. The justification is much more simple - you have to stop sometime and get a standard printed. Popular interfaces never stop evolving, adapting, and expanding to meet more uses than originally envisaged. An ongoing standardization effort began to extend the capabilities of SCSI-2. Some of the differences between SCSI-2 and SCSI-1 were that several options were removed from SCSI-1. Several new low-level requirements in SCSI-2 along with many options significantly enhancing SCSI. New command sets were added to SCSI-2 and all command sets were enhanced. Some of the features found in SCSI-2 include, wide SCSI with data transfer at bus widths of 16 and 32-bits. Fast SCSI achieves 10 MB/sec transfer rate with wider data paths of 16-, and 32-bits, rising up to 20 MB/sec and even 40 MB/sec. An active terminator lowering termination boosting system integrity. Several new commands were added, and diagnostics capabilities extended on the Read/Write buffer and Read/Write Long commands. Optional messages were added to negotiate wide transfers to support command queueing which extends up to 256 commands instead of one command. Sense keys and sense codes were formalized and extended. Figure 1 SCSI-3 Architecture Roadmap Accredited Standards Committee X3 has approved several SCSI-3 projects to enhance and restructure the SCSI-2 standard as shown in Figure 1. These projects (except fibre channel) are assigned to the X3T9.2 Task Group which developed this International Standard and the SCSI-1 standard. SCSI-3 is in fact the sum of a number of separate standards which are defined by separate groups. It is really a family of standards. SCSI was broken up from a single document into different layers and command sets. This was done to allow for different physical transport layers (like fibre channel and SSA) to be defined, and to allow for smaller "bite-sized" projects that may get done a little faster. Additional documents for the Fibre Channel are also meant to be included in the SCSI-3 framework. As all of this ongoing effort is considerably complex, it may cause changes in document structure and workgroups. The New Ultra SCSI Era Ultra SCSI is the next major performance advancement to the Small Computer System Interface and is a lower cost alternative to serial SCSI. Ultra SCSI boosts data transfer rates from the current limit of 10 MB/sec for Fast SCSI-2 to 20 MB/sec. It also doubles the Fast Wide SCSI-2 data transfer rate from 20 to 40 MB/sec. These performance improvements create the bandwidth necessary to support the data intensive applications to be used in the coming generations of servers, workstations, and high-end personal computers. The increased data transfer rates are attributed to the faster cycle times for data transfer and the arbitration of SCSI commands. Host systems and devices using Ultra SCSI will be able to negotiate optimal parameters for speed, width, offset, etc. These improvements and cycle times are primarily fueled by the higher speed of the new semiconductor technologies employed in SCSI chipsets. Since the internal drivers and the bulk of the drive circuitry and firmware remain the same, Ultra SCSI amounts to a low-cost, easy-to-integrate alternative for computer OEMs and system integrators who want to improve their products' I/O performance. Strategically, Ultra SCSI is the logical migration from SCSI to accommodate optimal system performance and ever-increasing demands by users for higher performance. While the infrastructure to support the Fibre Channel, SSA, and other serial interfaces are being developed, Ultra SCSI is the cost-effective solution for servers and workstations in the near term. Over time Ultra SCSI will appear in personal computers as integration costs decline. By effectively extending the life of the standard parallel interface SCSI provides a much needed stepping stone to the serial interface of the future. Some of the benefits of Ultra SCSI are: easy integration, reusability of current hardware and firmware, reusability of current test equipment, lower inventory risk. Due to Ultra SCSI's backward compatibility, it is a low-cost solution to higher performance. The first Ultra SCSI drives are already on the market. Drive manufacturers are working closely with host adapter companies to ensure compatibility across the SCSI bus. The first Ultra SCSI high capacity drives will feature buffer-to-host data transfer rates of up to 20 MB/sec for 8-bit Ultra SCSI implementations and up to 40 MB/sec for the 16-bit Wide Ultra SCSI option. These are double the speed found in Fast SCSI and Fast Wide SCSI high capacity drives, and follow-on activities are on the way which should increase these transfer rates even more. To obtain such performance improvements, and also maintain backward compatibility with Fast SCSI implementations, and to keep up with the internal disk data transfer rates, Ultra SCSI doubles the chip's internal clock speed (see Figure 2). This in turn doubles the millions of bytes that can be transferred per second. This approach is straightforward but it requires drive manufacturers to develop innovative solutions to considerable technical challenges in both digital and analog ASIC design. Figure 2 Ultra SCSI doubles the clock rate, doubling the throughput To ease the cost of migration from previous SCSI versions which have involved huge investments, Ultra SCSI is backward-compatible and uses the same physical environment. Ultra SCSI drives and systems can operate at the lower speeds of earlier SCSI versions for compatibility with older hardware. Cables connectors and terminators that support SCSI can support Ultra SCSI. More importantly, Ultra SCSI can be integrated without having to modify or change operating systems. Shorter Cables Equal Clean Signals Obtaining faster data transfers can pose challenges with analog chip design, which consists of I/O driver cells or transceivers. The inherent difficulty with a SCSI interface is that the cable which links the drive and the host system is a transmission line connection. The fast signal transitions or slew rates associated with high-performance analog circuit designs often create transmission-line effects such as ringing, overshoot, and undershoot which cause signal quality problems, and can interfere with the interface chip's transceiver performance. The traditional solution, which is to significantly slow down slew rates, wouldn't work in this case because it would also prevent the interface chip from meeting the timing margins needed for Ultra SCSI data transfer speeds (see Table 1). Instead, Ultra SCSI's developers drew upon their experience with potential transmission-line effects in designing earlier Fast SCSI drives: Earlier work had shown that shortening the cable length lessened the transmission line effects and therefore reduced signal degradation to acceptable levels. Thus Ultra SCSI drives today require a standard 1.5m limit on cable length, though work is continuing to increase that length in the near future. The 1.5m cable length makes the trade-off between slowing down the skew rates and meeting overall timing requirements easier to manage. For in-the-box hard drive configurations as well as many outside-the-box configurations such as desktop system configurations to a CD-ROM, a 1.5m cable proves more than adequate. For machine room or server environments in which a large number of devices need to be connected, however, that length limitation can be restrictive. In these cases the alternative is to use differential Ultra SCSI drives, which provide greater noise immunity than the more widely used single-ended SCSI products and can support cable lengths of up to 25m. In differential SCSI a pair of wires carry each signal that needs to be sent across the bus. The first pair carries the same type of signal the single-ended SCSI carries, and the second pair carries its logical inversion. The receiver takes the differential of the pair thus the name differential which makes it less susceptible to noise and allows for greater cable length. Although differential SCSI drives are more expensive than single-ended SCSI drives their price tags can be easily cost justified for server and other high-end drive applications. Table 1 SCSI Bus Timing Values New proposed devices referred to as bus extenders (which act as repeaters like those used in communications) let the host be located as much as 6 meters from the first SCSI device. These extenders can convert the 1.5-meter mini-bus of Ultra SCSI into the stretch limo of buses with an overall length of 7.5 meters. For many multidrive systems the 1.5-meter limitation on the back plane connection presents no problem, because the drives are located in a single box. Bus extenders can be used to create systems in the form of building blocks, giving designers more flexibility. And the cost is trivial compared to its benefits: single ended bus extenders are projected to be priced at (cost) less than $10. The proliferation of disk drives in servers, RAID, and video-on-demand systems can stress SCSI's logical connectivity. Under the existing SCSI arbitration protocol, the logical number of devices that can be addresses is limited to the number of available data lines: Eight devices including the host for narrow SCSI, 16 devices for Wide SCSI. But there is a growing need for more connections. One means of addressing this need is a two-phase arbitration technique called dual-phase protocol, currently under development in ANSI's SCSI committee. This addressing scheme is accomplished through delivery of the higher bits of data of a SCSI-device address in the first phase. The result of the first phase of addressing determines which devices are eligible to participate in the second round. The lower bits are then used to determine the selected device. With this scheme, the number of addressable devices is equal to the square of the number of data lines, or up to 256 devices with a 16-bit interface, or, for those who demand the ultimate limit, up to 1,024 devices with a 32-bit parallel interface. Comparing the Multiple Paths to Higher Performance There is a great demand in the market today for a giant step forward toward a very fast, highly efficient hard disk interface. Which interface will provide a commuter lane for data on the highway where the rush hour never ends? The competing serial interfaces, FC-AL and SSA have received a great deal of attention by the press because they are relatively new, with promises and dramatic results. In fact, the publicity gained by these serial interfaces tend to overshadow the impressive gains available in the familiar territory of parallel SCSI. While the new serial interfaces do offer immediate improvements, parallel SCSI continues to offer evolutionary growth, and an alternate route to promising performance. Throughout the history of innovation we've seen that transitioning from an existing technology to a new technology has often been less rapid than anticipated. Invariably, an older technology responds to the same market pressures that created the new one. This results in enhancements which give it a whole new life. In fact, in many cases the old technology ends up being better positioned to respond to the new market demands. For example, in the magnetic world, thin film heads were envisaged to revolutionize the industry when they were first introduced six years ago. They indeed boasted higher signal performance, but simultaneously older technology Ferrite heads moved forward offering improved MIG (metal-in-gap) designs at lower costs. Thus, while serial interfaces prove higher performance, some improvements could make the parallel interface a viable, cost-effective choice. The SCSI interface has come a long way with Ultra SCSI transfer rates of up to 40 MB/sec. and increased connectivity for up to 16 devices to be connected to the bus, as opposed to seven devices when SCSI was first introduced more than a decade ago. In a year or two transfer rates are expected to move up to 80 MB/sec with Ultra SCSI, and possibly to 160 MB/sec with the 16-bit version. With today's high density connectors and direct attachment to the backplane, this version will actually be easier to configure and more reliable than the original 5 MB/sec SCSI. Parallel SCSI has quite a few advantages over the new serial interfaces, such as the software for I/O in the operating system, drivers and subsystem logical addressing, firmware, and a fortune in hardware that backward compatibility can save from the junkyard. Parallel SCSI system design, manufacturing, support, sources, experience and user familiarity seems a better choice. There are applications for which even these expanded capabilities don't suffice. In large scale applications designers resort to multitudes of host adapters and scary masses of cabling, but the next step solution should ensure range and load carrying capabilities needed far into the future. In comparing FC-AL, Ultra SCSI and SSA we find that FC-AL speed leaves other contenders in the dust. With the turbo charging of dual-porting, FC-AL can transfer data at 200 MB/sec, with acceleration up to 400 MB/sec not far off into the future. This bandwidth gives FC-AL a distinct edge when the goal is to interconnect a large number of drives. Individual drives will soon offer sustained data rates of over 10 MB/sec, so a 200 MB interface bandwidth is definitely not overkill in large systems. FC-AL covers distance spanning more than 30 meters, using copper or coax and up to 10 kilometers with fiber optic. The requirement for more storage has presented new challenges such as combining incredible numbers of drives in arrays, and providing operation without interruption. FC-AL offers viable solutions to these challenges. Both FC-AL and SSA assure high fault tolerance dual porting. Both parallel SCSI and FC-AL permit hot plugging, i.e. replacement of drives without disrupting disk array operation. Though FC-AL is not expected to appear in products until 1996, ten years from now it is expected to be the established serial interface. Any practical implementation of SSA and FC-AL is probably years away, since most users are comfortable with SCSI. And don't be surprised if a decade from now people are still debating the decline of parallel SCSI. Designers and system integrators specifying disk drive interfaces for computers are faced with a choice between evolutionary and revolutionary technologies. One path leads to adequate performance improvements with minimal integration efforts and no additional costs. The other path promises great performance gains in exchange for additional implementation efforts and costs. For many designers, the best path is to take full advantage of the performance improvements offered by the evolutionary technology. Other will forge ahead, recognizing the advantages of the revolutionary path, and choose to be the early adopters of the new interface. In the disk drive industry today, the evolutionary technology is represented by the Ultra SCSI interface, and the revolutionary technology is represented by the serial interfaces.