Historically, RISC-based servers have dominated the eight-way and larger server categories. RISC-based servers are so named because their processors use a design known as reduced instruction set computer (RISC). RISC processors use customized instruction sets tailored to the tasks the processor is designed to do, separate instruction and data caches, instruction pipelining, and superscalar operation to permit multiple instructions to be performed at the same time. At one time, the differences in performance between RISC and CISC designs such as x86 processors was profound. However, starting with sixth-generation processors such as the Pentium Pro and its many descendents, x86 processors have adopted many RISC techniques to improve performance. The EPIC processor architecture used by the Itanium family has elements of both RISC and CISC designs. At the same time, the clock speeds of Itanium 2 processors have matched those offered by the fastest RISC processors, and x86 processors offer considerably faster clock speeds. Note When gauging processor performance, pure clock speed is only one factor in gauging a chip's overall performance. As a result of the performance gains in x86-based chips, several families of RISC-based processors, including the Alpha, PA-RISC, and MIPS R1xxxx will be phased out in the next year or two. While RISC technologies dominate embedded computing, RISC is no longer the undisputed champion of network servers. For this reason, much of this book is focused on Intel and AMD-based system designs. The following sections provide additional details about major RISC-based sever processor families. AlphaThe Alpha processor family is often referred to as the DEC Alpha because it was originally developed by one-time powerhouse Digital Equipment Corporation in 1992 as the Alpha 21064. The code name EV applied to various Alpha designs is short for "extended VAX," as the Alpha was originally intended to run DEC's VAX operating system. Major features of the Alpha architecture include the following:
The success of the AMD Athlon family indicates the fundamental strength of the Alpha design as well as its biggest weakness: The Alpha uses a short, very efficient pipeline, which makes it able to handle frequent changes in instructions and data. However, the shorter the pipeline, the harder it is to scale the processor to higher clock speeds. Consequently, while the Alpha processor offered very competitive clock speeds in the late 1990s, it fell behind compared to other RISC designs such as PowerPC or x86 and Itanium 2based chips. However, the major reason for Hewlett-Packard's decision to stop selling AlphaServers in 2006 is that Hewlett-Packard has three 64-bit server platforms. Hewlett-Packard had already decided to phase out its PA-RISC 8x00-based platforms, such as the Hewlett-Packard 9000 and Superdome series, because of its co-development with Intel of the Itanium processor family. When Hewlett-Packard purchased Compaq in 2001, Hewlett-Packard found itself with another unwanted RISC processor family that Compaq had already decided to abandon by 2004. In effect, Hewlett-Packard's purchase of Compaq gave the Compaq/Hewlett-Packard AlphaServer line a two-year reprieve that is now coming to an end. The latest AlphaServers use the 21364 processor. Compaq, and now Hewlett-Packard, have offered AlphaServers in four product lines, as detailed in Table 2.33:
Hewlett-Packard is not the only source for AlphaServers or workstations, however. Microway (www.microway.com) offers customized Alpha-powered systems that use the 21264 Alpha processor. Microway took over support and warranty service for nonHewlett-Packard Alpha systems in 2001 when API (a joint venture between Compaq and Samsung, started in 1998 to manufacture and support Alpha processors and systems) quit the Alpha business.
PA-RISC 8xxxHewlett-Packard's PA-RISC 8000, introduced in 1996, was the first 64-bit version of its PA-RISC processor line, which was first used in the original 32-bit form in the Hewlett-Packard 3000 minicomputer line starting around 1989. The PA-RISC 8xxx family has undergone only relatively minor design changes since it was first introduced. The major features of the PA-RISC 8000 include the following:
Improvements added to later versions include the following:
Hewlett-Packard uses the Hewlett-Packard 9000 brand for its PA-RISCbased server product line. Current Hewlett-Packard 9000 servers are listed in Table 2.34
For more information about current Hewlett-Packard servers, see the Hewlett-Packard website, at www.hp.com. For more information about PA-RISC hardware and software, past and present, see the OpenPA website, at www.openpa.net.
MIPS R1xxxxAlthough the MIPS R-series is associated with SGI, it was actually developed by MIPS Computer Systems in 1985. SGI bought MIPS in 1991 after MIPS ran into financial trouble during the development of the world's first 64-bit microprocessor, the R4000. After several other 64-bit processors were developed in various series, the R10000 processor was introduced in 1995 as an improved version of the R8000 and was used by many companies besides SGI. The R10000 processor has the following major features:
SGI had planned to switch from MIPS to Itanium processors in the late 1990s, but when the Itanium was delayed, SGI was forced to continue to develop the R10000 design with subsequent versions:
Currently, SGI uses the MIPS R16000 in its SGI Origin 350 technical servers and Origin 3000 supercomputers. These systems, which run SGI's IRIX implementation of UNIX, offer highly modular and expandable designs and more closely resemble server clusters in a box than a conventional multiprocessor server. For more information, visit www.sgi.com/products/servers/origin/. The other major vendor currently offering MIPS-based servers is Hewlett-Packard. Its NonStop S-series line of high-availability servers uses matched pairs of R12000, R14000, or R16000 processors running in lockstep, depending on the model. See www.hp.com for details. Both Hewlett-Packard and SGI are expected to eventually move completely to Itanium-class processors for their entire product lines. Most of SGI's current servers have already moved to Itanium 2, and Hewlett-Packard is transitioning its AlphaServer and Hewlett-Packard 9000 (PA-RISC) users to Itanium 2 systems. Hewlett-Packard also offers a line of NonStop servers that use Itanium 2 processors.
Power ArchitectureThe most successful RISC processor vendor appears to be IBM. It continues to develop two distinct families of RISC-based processors: the PowerPC, discussed in the following section, and the PowerPC's big brother, Power Architecture (discussed in this section). Power Architecture processors are known as Powerx processors, ranging from the original 32-bit Power1 (RIOS) processor used by the RS/6000 line of servers and computers in 1990 to the current Power5+. Early Power processors used multichip designs. However, the Power2 SuperChip (also known as the P2SC) was the first single-chip Power Architecture processor, and it was the last to feature 32-bit operation. Its major features included the following:
The Power3 added the following:
The Power4 added the following:
The Power5 adds the following:
The Power5+ is a die-shrunk version of the Power5. Current eServer Open Power and System p5 IBM servers that use Power chips are listed in Table 2.35. Power chips are also used in other product lines, including the iSeries integrated business system.
PowerPCThe PowerPC chip family, originally introduced in 1992, is based on the Power Architecture Power family. PowerPC chips were originally developed as single-chip versions of the multichip Power1 processor that could be used in other IBM products in a cooperative effort between IBM, Apple, and Motorola. The 32-bit PowerPC chip 604 was used in several servers manufactured by Apple, Motorola, and IBM, among others. Windows NT was available for early PowerPC systems but was quickly discontinued. The 604's major features included the following:
The 604e added larger instruction and data caches. The 740 and 750 are members of the G3 family and were the first commercially successful 64-bit PowerPC chips:
Both also support 32-bit PowerPC code and were used in Apple PowerMac G3 servers. The 7400 processor (also known as the G4) was manufactured by Freescale (Motorola's former processor division) and was co-designed by Motorola and Apple. IBM did not participate in its design. The G4/7400's major features include the following:
The G4/7400 was used in Apple PowerMac servers. The latest PowerPC processor family, the 970, is also known as the G5. It was developed by IBM. Major features of the G5/970 include the following:
The 970FX is a die-shrunk version of the 970. The 970MP is a dual-core version. All 970-family processors have been used in Apple PowerMac G5 xServe servers. The 970 is also used by IBM's BladeCenter JS20 server blade. The long-term future of PowerPC chips at the server level is questionable because Apple has switched to Intel Core processors for its latest Macintosh products. However, PowerPC chips continue to be popular in video game and embedded applications, and IBM may continue to use them in its own server products. For more information about PowerPC processors, see the IBM website, at www.ibm.com, and the Freescale website, at www.freescale.com. For more information about PowerPC-based servers, see the Apple website, at www.apple.com.
|