Printer Friendly

WHO NEEDS 64 BITS?

The computer industry is in the throes of a transition to 64--bit computing that will expand digital horizons. In electronic CAD, 64--bit systems are already making possible the analysis of more complex integrated circuits. In mechanical CAD, 64--bit computing is a prerequisite - though not the only one - for ushering in the day when engineers can interact with very large assemblies and not just fly through graphical representations stripped of their parametric features. Optimists about 64--bit computing, which has been sitting in the on--deck circle for more than five years, take heart that the limits of 32--bit computing are becoming increasingly obvious. DRAM prices are falling to the point that multigigabyte main memory pools are no longer unheard of in servers and even in the latest workstations. Memory chips (and the busses over which they communicate with CPUs) also are getting a lot faster. The two--gigabyte limit on per--process memory imposed by 32--bit operating systems looks as constraining as a pair of undersized shoes. More to the point, localized disk arrays are now available that can warehouse hundreds of gigabytes of data. Storage--area networks with dozens to hundreds of terabytes can be tied to workstations and servers (see the September and October CAD Reports). Through use of RAID and high--speed SCSI or fiber--channel connections, this data can be transported to and from workstations and servers more quickly than ever before. Jamming this data into the tiny file--system and memory areas permitted by 32--bit operating systems is like using a soda straw to fill a swimming pool. Sun Microsystems's Robert Novak claims that the increasingly affordable and powerful data--storage systems combined with fast crossbar switches for moving that data is already giving an advantage to 64--bit platforms capable of loading huge models and databases. He says the biggest pressure from Sun users for 64--bit computing is coming in electronic design automation from engineers trying to model ever bigger chips. Other areas likely to benefit first are computer aided engineering and enterprise--transaction processing. The day when 64--bit computing enters the technical mainstream is at least a few years off. Although a lot of the groundwork has been put in place, plenty still needs to be done. Most importantly, demand for the benefits will have to come from more end users, most of whom have yet to exhaust the promise of 32--bit computing. But lack of demand hasn't stopped computer vendors from heralding the advance to 64--bit computing. To decide when and how 64--bit systems might help your company it helps to have a firm grasp of some key concepts. What Does It Mean? Sixty--four--bit computing is a general term for a number of ideas. These include how variables are defined and manipulated, hardware considerations such as the bit--widths of CPU registers and data busses, and system--software issues. Higher Precision The more bits you use to represent a number, the greater its precision will be. For this reason, 64--bit systems have greater inherent accuracy than 32--bit systems. But in technical niches, the improved precision that comes with 64--bit data variables predates 64--bit CPUs, busses, and software. Thirty--two--bit applications can specify that values be represented in double--precision mode. This technique allows two 32--bit data registers to be set aside to hold 64--bit numbers. Squeezing such big numbers into CPUs that have 32--bit data structures is somewhat inefficient. It entails extra overhead such as having to manage overflows and underflows across the 32nd and 33rd bits of the variable. But overall this additional work has been a fairly minor concern of software vendors because the performance penalty has been minimal. In terms of precision, 64 bits is old hat, particularly in engineering programs where rounding errors can be disastrous. Larger File Systems A 64--bit operating system permits disk drives to be partitioned into larger logical units (called volumes) than 32--bit systems. Strictly speaking, a 32--bit system has a four--gigabyte limit for file volumes, though most 32--bit operating systems have been revamped to allow for larger file systems. Even Microsoft's PC operating system Windows 98 now has a four--terabyte volume limit. The typical 64--bit OS today allows for file systems that can hold up to 16 terabytes. SGI's 64--bit Irix can handle nine million terabytes, a spokesperson says. 64--bit Hardware Computer central processors are frequently classified by the size of their data registers. Early minicomputers and PCs had 16--bit registers. Early IBM mainframes employed 24--bit registers. Registers are internal storage locations that hold data, memory addresses, and program instructions for the processor execution units. In a 32--bit CPU, each register has 32 individual cells that can store one 32--bit number. A 64--bit CPU works with registers that have 64 such places. In reality, the internal CPU registers may have more places than would be indicated by the CPU bit--size. These extra places are used to designate different modes of operation into which the CPU may switch or to hold communications settings and the like. But it's the standard size of the data chunks manipulated internally that gives the CPU its name as a 16--bit, 32--bit, or 64--bit microprocessor. Communication pathways (or busses) are also defined by their widths in bits. A 32--bit bus is composed of 32 signal lines; a 64--bit bus has 64. Sometimes there is not a direct correlation between the bus width and the register widths of the CPU. Some 64--bit Alpha CPUs are fed by 128--bit memory busses. These wider busses can move 64--bit data chunks twice as fast as systems with 64--bit busses running at the same clock speeds. The UltraSPARC CPUs in Sun's Ultra 60 workstations are fed by 578--bit memory busses. In addition to registers and busses, the sets of chips that control the communication lines to and from the CPU and interact with cache and main memory also come into play. They have to be upgraded to access the additional memory addressable by 64--bit software. 64--bit System Software To take full advantage of 64--bit microprocessors, operating systems and other system software must be revised. Among these revisions, the OS must be able to control a lot more memory. Other libraries of procedures governing a host of activities such as handling disk I/O (input/output) or responding to graphics calls have to be revamped. Of all of these system--software issues, the toughest nut to crack is developing a 64--bit memory model. Technically speaking, an operating system keeps tabs of a model by storing each byte in a table known as an array. Each entry in the array refers to the unique address (or location) in virtual memory (VM), where the data for an active process is held. VM is a combination of main memory and (when there is not enough DRAM) hard--disk swap space. The maximum number of address entries in one of these arrays is fixed and is in part a function of the operating system's bit size. A 32--bit operating system has a theoretical limit of about four billion entries, though most existing 32--bit systems can manage only about two billion bytes per process. Currently existing 64--bit operating systems can manage arrays ranging from four trillion to 16 trillion entries, with the theoretical limit about a million times more than that. Two Approaches: Lots of Confusion Authors of operating systems have taken two basic approaches in putting all of this groundwork in place. Digital Equipment launched the 64--bit revolution in one bold stroke. In 1993, it introduced the 64--bit operating system DEC OSF/1 to go with its new 64--bit Alpha CPU. Releasing the 64--bit hardware and 64--bit operating system together helped Digital gain a performance lead over rivals and simplified the work of building the new systems. The disadvantage was that applications that ran on Digital's earlier Ultrix flavor of Unix and DEC's older MIPS--based platforms that preceded the Alpha couldn't run without modification on the new architecture. The breaking of backward compatibility in this way is something other vendors have been less willing to risk. As an alternative, the other Unix vendors have taken a more gradualist approach, in all cases extending existing operating systems rather than writing new ones. The 64--bit operating systems from SGI, HP, IBM, and Sun also have 32--bit libraries so they can execute legacy 32--bit code. Their 64--bit CPUs come equipped with mechanisms to recognize 32--bit instructions, shift modes, and perform the right calculation. Here is a brief outline of the 64--bit migration of the leading Unix vendors: Silicon Graphics introduced a 64--bit CPU (the MIPS R8000) along with a 64--bit version of its Irix operating system (Irix 6.0) in 1994. But even today, many of SGI's customers continue working with 32--bit software, using 64--bit Irix 6.4 and 6.5 and the prevalent MIPS R10000, a 64--bit CPU. Sun Microsystems introduced the 64--bit UltraSPARC CPU architecture in 1995, but Solaris 2.7, with 64--bit addressing, wasn't unveiled until late October of last year. In between, Sun incorporated several 64--bit features into earlier versions of Solaris, including large file systems. Hewlett--Packard took a similar approach to Sun's. The PA--8000, which came out in 1996, was HP's first 64--bit CPU. The PA--8500, the latest PA--RISC chip, is a third--generation 64--bit microprocessor that is currently in volume production. Sixty--four--bit HP--UX 11.0 came out in November 1997 for HP Unix servers and about six months later for PA--RISC workstations. Meanwhile, HP has continued adding to its 32--bit PA--7000 generation and upgrading HP--UX 10.X, its most advanced 32--bit version of the operating system. In 1997, IBM debuted AIX 4.3, a 64--bit operating system, and the RS64, a special 64--bit version of the PowerPC microprocessor developed for the RS/6000 S70 server and the AS/400 minicomputer line. Recently, IBM started shipping its first computers built around the POWER3, a 64--bit RS/6000 CPU (see the November 1998 CAD Report). IBM is recommending the latest version of its Unix, which is AIX 4.3.2 for both 32--bit PowerPC 604e workstations, as well as for the POWER3. These gradualist strategies allow plenty of time to get out the bugs, to educate customers about 64--bit computing, and to drum up support from application vendors. However, each time a new 64--bit feature has been introduced, the advertising material makes it sound like the whole migration has been completed. It's sort of like saying a jigsaw puzzle is finished every time an additional piece is fit. Sixty--Four--Bit Applications The real payoff from 64--bit computing doesn't come until application vendors rewrite their software to take advantage of its features. In this regard, 64--bit computing is making a big splash in high--end transaction processing, where supercomputers are used to manipulate very large databases in "mainframe replacement" applications. But when will 64--bit computing take over CAD markets? Most mechanical CAD vendors have largely ignored the shift to 64--bit systems. All of the pressure right now is coming from hardware vendors, not end users, say spokespeople for a couple of important MCAD companies. Secondly, developing and supporting 64--bit versions of existing software will be costly, they say. Part of the problem, according to Christophe Mathieu of Dassault Systemes, is the old bugaboo of assuring backward compatibility for legacy 32--bit code and not making obsolete the 32--bit hardware that most of the customers are still using. Bob Brandenstein of Unigraphics Solutions agrees. With 32--bit and 64--bit versions of the same software, "You have to have machines in both flavors for your developers. You have to certify the machines in both flavors for your customers. And when a customer calls, you have to ask what flavor they are using. Even if it were a trivial concern (to develop a new 64--bit version), and it's not, there's a fair amount of overhead for your organization to support." Brandenstein and Mathieu both say that there has been little demand from their customers for a 64--bit version of their products. Validating this point is that Digital Unix can handle up to four terabytes of virtual memory. But this descendant of the oldest commercially available 64--bit operating system defaults to one gigabyte in standard installations. Why? Typical users don't need more than a gigabyte of virtual memory, says a Compaq spokesperson. The default setting helps optimize the operating system for the typical end user, he said. Brandenstein said that the reason there isn't more enthusiasm for 64--bit computing is that today's top processors and graphics adapters still aren't powerful enough to work with solid assembly models big enough to surpass 32--bit's two--gigabyte limit. He said that Unigraphics has experimented with several high--end workstations containing 768 megabytes of DRAM and found that the performance and interactivity on models larger than 512 megabytes was too slow. However, Brandenstein said he is hopeful that within a year or two faster hardware will make modeling these large designs feasible. Mathieu said that some of Dassault's biggest customers such as Boeing, Chrysler, Bombardier, and Dassault Aircraft are approaching the limits imposed by 32--bit computing. When customers need it, Dassault will make the change, and the shift is part of the company's long--range plans, he said. Mathieu added that someday Dassault Systemes might develop a 64--bit version of CATIA version five. However, because version five supports both Unix and Windows 2000 (a.k.a. NT), it is unlikely that Dassault Systemes will make its move until Intel's newly developed IA--64 architecture catches on and Microsoft develops a 64--bit version of Windows. Dassault Systemes's dilemma points to a further drag on the arrival of the 64--bit era. Currently, makers of CAD software are focused on the Windows NT takeover. While NT on Intel has been a boon on the economic front, it has slowed the switch to 64--bit computing. Intel's first 64--bit CPU, code--named Merced, is scheduled for release in the middle of the year 2000. Microsoft will incorporate a 64--bit file system in Windows 2000 to be timed with the Merced release. Because Windows won't be ready to take advantage of many of Merced's 64--bit features, particularly larger memory--handling capabilities, Intel has been working more closely than ever with Unix vendors in preparation for the introduction of its IA--64 architecture on which Merced is based. HP, Sun, SGI, the Santa Cruz Operation (SCO), and Compaq, following its Digital acquisition, are competing to define a Unix standard for IA--64. (See the February 1998 CAD Report.) Lots of Head Room With or without Microsoft's cooperation, the desire of technical users to work with larger models will eventually create demand for 64--bit systems. And fortunately, the transition to 64 bits is likely to be the last such migration any of us will live to see. At least theoretically, a 64--bit operating system can store data in up to 264 memory locations. This would make it possible to directly address 16 "exabytes" of DRAM, which is greater than 18.446 X 1018. Even at commodity prices (say a dollar a megabyte), outfitting a system with that much DRAM would cost $17,592 trillion. To put this in perspective, the current U.S. national debt is about $5.4 trillion. Still, it's comforting to hear about something that makes the debt look small and also to know that it will be a very long time before we start hearing about the shift to 128--bit computing.
COPYRIGHT 1999 Worldwide Videotex
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1999 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Publication:CAD/CAM Update
Date:Feb 1, 1999
Words:2530
Previous Article:RECOVERING LOST PRODUCTIVITY.
Next Article:THE OBLIGATORY Y2K ARTICLE.


Related Articles
Digital Voice and Multiplexing.
IPv6 to the rescue?
SUN SOLARIS UP ON MERCED SDK.
FUJITSU INTRODUCES 64-MBIT SYNCHRONOUS DRAM FOR PC133.
AMD Plots Easier Route to 64-bits With Sledgehammer.
NORTEL NETWORKS: NEW SOLUTION FOR FASTER UMTS WIRELESS DATA SERVICE.
AMD's Opteron aims at Itanium in 64-bit battle.
Protect your network at the speed of light.
Ag-machine maker depends on drill sharpeners.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters