UNIX THREATENED BY MICROSOFT AND INTEL
Sun's spectacular rise and descent was due largely to how Sun responded to changing market conditions. Understanding those market conditions is therefore necessary for understanding how Sun's fortune could have altered so dramatically.
In the mid-1990s, before Sun's amazing ascent, the entire Unix market was seen as gravely endangered by threats from below. The Intel/Microsoft juggernaut had recently finished destroying its competition in the desktop space, and began eyeing the enterprise market for takeover. This takeover was to have two parts. First, Intel would introduce a new high-end processor to compete against the RISC processors common in the enterprise space. Second, Microsoft would start pushing its new version of Windows (called "NT") as an alternative to expensive and proprietary Unix offerings.
The first part of the takeover, Intel's new processor, would employ a fundamentally new architecture that would be superior to the RISC designs used by the high-end Unix vendors. The main idea behind this new architecture (called "EPIC") was: incoming instructions would be grouped into bundles, and the compiler would decide (at compile time) which instructions would be executed in parallel and how they would be scheduled. This was an extension of the RISC philosophy of shifting work from the processor to the compiler. Since the compiler could spend hours analyzing the instruction stream, it could extract far more parallelism than the RISC processors could extract on-the-fly. And since the logic for determining instruction scheduling and parallelism was no longer embedded in the processor, but in the compiler instead, a huge number of transistors could be liberated for performance-enhancing features like more units, bigger caches, and so on. Thus, EPIC promised to be a revolution as great as the original RISC revolution; EPIC promised vastly greater performance, on smaller die sizes, and for less money.
The software side of the takeover was supplied by Microsoft, in its new operating system called "Windows NT." This new operating system was based on a modern, pre-emptively multitasking, protected memory kernel -- just like the kernels contained in unix operating systems. Thus, it would provide all the reliability and stability of Unix OSes. But NT, unlike Unix, would be common, inexpensive, and easy to use, greatly reducing the administrative expenses and total cost of ownership for organizations that deployed it.
The combination of Microsoft's new operating system, and Intel's new processor, would constitute a crushing blow to the Unix vendors. Combined, Microsoft and Intel would bring high volume, low prices, common components, and superior performance to the weak, fractured, low-volume commercial Unix market. The Microsoft/Intel juggernaut would roll over the Unix vendors, in the same way it did Apple and all its desktop competitors. It became common wisdom, in the mid-1990s, that RISC and Unix were doomed.
This sentiment of doom was repeated by the industry analysts. The analysts agreed that the Intel/Microsoft juggernaut was unstoppable and would bulldoze over proprietary Unix offerings. Soon, the analysts said, Unix would start gradually fading away, and RISC would be destroyed. One very prominent analyst wrote, in 1998, a report on EPIC versus its RISC competitors, and concluded that if the market were a chess match then checkmate had already been declared and EPIC had already won. This was surprising, given that no EPIC processor had yet been released and none was going to be released for three more years. But the other analysts agreed. According to the "common wisdom" of the day, the Microsoft/Intel assault was as good as won, even before it had begun in earnest.
THE UNIX VENDORS RESPOND TO THE MICROSOFT/INTEL THREAT
For the most part, the commercial Unix market agreed with the consensus that it was doomed. Most of the commercial Unix vendors quaked in terror at the prospect of a Microsoft/Intel onslaught, and several Unix vendors reacted by announcing the demise of their own platforms, thereby shooting themselves in the head before Micorosoft and Intel could get to them. SGI announced (in 1997) that it would cancel its upcoming CPUs, cease development of its own processor family, and transition its users to EPIC. HP announced that it would cease development of its PA/RISC processor family and would transition its users to EPIC as well. IBM decided to "hedge its bets" by deploying both EPIC systems and its own POWER systems concurrently, but downplaying the importance of its own POWER systems. And the CEOs of both SGI and HP seriously insinuated that they might cancel their respective Unix OSes (in the distant future) and transition their users gradually to Windows NT.
Sun was the only company which didn't participate in this killing of its own children. Instead, Sun went in the exact opposite direction. Instead of announcing the imminent demise of its processor family, Sun announced that it was designing the next generation of those processors. Instead of insinuating that the Solaris OS might be cancelled, and its users eventually migrated to NT, Sun aggressively promoted the Solaris operating system and derided NT. And although Sun initially dabbled with the idea of building EPIC boxes, alongside its traditional RISC boxes, Sun quickly abandoned that idea and decided to commit all its resources entirely to its own platform. Thus, Sun redoubled its efforts, introduced new models, and even went so far as to purchase (from SGI/Cray) the designs to a massive new Unix server that would be the largest and most expensive Unix server ever. If Unix was dead, Sun apparently was unwilling to believe it.
Because of this unswerving commitment, Sun quickly captured most of the Unix market from its rivals. This wasn't due to technological superiority, but to the fact that Sun appeared to be the only vendor not abandoning Unix. Therefore, Sun won the Unix market because the other unix vendors, believing the market to be doomed, essentially abandoned it. The other unix vendors (in essence) voluntarily ceded the market to Sun.
THE INTEL/MICROSOFT JUGGERNAUT STALLS
The analysts had been in agreement about the imminent demise of Unix. But the difficulty with analysts' predictions is that they're generally either obvious ("computers will get faster next year") or completely wrong. The more analysts agree on some prediction, the more likely they are to be wrong. In predicting the death of commercial Unix/RISC systems, the analysts were virtually unanimous, and the analysts were almost totally wrong.
Cracks started to appear in the Intel side of the takeover when its upcoming Merced processor was dreadfully late, and started falling further and further behind schedule. Schedule slippage was soon accompanied by warnings, issued by Intel, that the new processor's performance would be well below what had been promised. Intel began downplaying the importance of the new processor, stating openly that the subsequent processor ("McKinley") would really be much better. When the new processor was eventually released, its performance, far from being revolutionary, was significantly below currently available RISC offerings on commercial workloads. This tepid performance, combined with a lack of available software, led to disastrous marketplace results. After Merced had been selling for more than a year, sales figures indicated that only single-digit thousands had been sold. Intel's terrifying onslaught had failed to take more than 0.1% of the server processor market.
Microsoft's portion of the assault, the "Windows NT" operating system, was enormous, complicated, and far too immature for enterprise computing. The combination of immaturity and compexity led to innumerable bugs, which Microsoft was incapable of repairing in a timely manner. Soon the buggy OS gained a reputation for serious unreliability, making it inadequate for anything but low-end non-mission-critical servers. Microsoft's marketing machine only exacerbated this problem by convincing credulous managers to convert to NT when NT was not ready, leading to disastrous failures and "re-conversions" back to Unix.
Thus, Intel's EPIC had failed completely, and Microsoft's NT had penetrated only the fringes of enterprise computing. The analysts were proved wrong. The Juggernaut had stalled. The initial foray into the enterprise market had failed.
UNIX EXPERIENCES A RENAISSANCE, TECH ECONOMY EXPLODES, SUN BLOOMS
The failure of the Microsoft/Intel assault left Unix without any serious contenders in the enterprise space. This caused, starting in 1999, something like a Renaissance of Unix. All of the sudden, Unix was seen as being cool again, even "cutting edge," which was surprising for a technology that was 25 years old. Instead of converting away from Unix, as had been so widely predicted, companies began converting to it. And since Sun had picked up so much of the Unix market as its competitors abandoned it, Sun became the natural beneficiary of the Unix Renaissance. Sun's sales and earnings surged.
Adding to Sun's fortune was the earlier release (in 1997) of its new server, the E10000, codenamed "Sunfire." Based on a design Sun had purchased from SGI (which had in turn acquired the design from Cray), this server had 64 tightly coupled CPUs in a huge SMP configuration that was larger and more expensive than any previous Unix machine. When the E10000 was announced, and Unix was presumed dead, a prominent analyst predicted that Sun would sell, at most, a "few dozen" of them. Instead it sold briskly and differentiated Sun from its competitors, since nobody else had anything like it. Sun's E10000 took the top ranking in the heavily followed tpc-c benchmark, being the first server to top 100,000 transactions/second.
But the real boon to Sun was the massive explosion of the tech economy. The Internet suddenly took off like a rocket, and following the Internet boom was the greatest speculative investment boom in the history of mankind. Trillions of dollars flowed into "dot coms" and other technology startups, and all these newfound companies, especially the internet-related companies, required servers. Sun, being the dominant maker of the now well-regarded Unix, seemed like the natural choice. Sun became regarded as the "standard" choice in servers, and as small companies and startups frittered away trillions of dollars of investor money, Sun happily sold them large amounts of expensive equipment. Several companies bought Sun's enormous multimillion-dollar E10000s as a matter of pride and "showing off."
This tech boom, combined with Sun's recent dominance in the Unix market, let to a skyrocketing of Sun's sales and earnings. Sun's stock price jumped to become 30 times higher than in had been. Sun became a $200 billion company that was twice as large as its formerly much larger foe, HP, and that was approaching IBM in market capitalization. Sun was seen as the vendor supplying computers to power the New Digital Age. Sun was on top of the world.
Thus, Sun's fantastic ascent was largely due to three factors: first, the abandonment of the Unix market by Sun's competitors; second, the technological failures of the Intel/Microsoft juggernaut; and third, the Renaissance of Unix with an accompanying tech economy boom. Each of these factors contributed greatly to Sun's ascent, but each of these factors was, in the end, transient.
The first factor to give way was the tech economy boom. Speculative booms have a history of collapsing as dramatically as they had risen, and the "dot com" boom was no exception. When the collapse came, the NASDAQ stock market lost more than 75% of its value, and almost all of the previously high-flying "dot coms" quickly were driven into bankruptcy. This collapse of the "dot coms" was a special difficulty for Sun, since almost all of them had been Sun customers, and Sun was heavily reliant upon them for revenue. So, as the dot coms went bankrupt, the market for Sun's equipment sharply contracted. This contraction was exacerbated by the fact that these bankrupt startups dumped piles of their mostly unused Sun hardware back onto the hardware market, at a fraction of the original price, creating a glut of Sun equipment.
But the tech economy collapse was to be only the first of Sun's problems. Micorosft's and Intel's initial forays into the enterprise space might have been disasters, but those two companies are endlessly tenacious and have almost unlimited resources to devote to new attempts. So, of course, they would try again. Microsoft released a new version of its NT operating system (now renamed "Windows 2000"), and the new version proved to be vastly superior to the old. And Intel released a new version of its Itanium processor (called "Itanium 2"), and its strong performance surprised almost everyone, offering benchmark scores similar to the best of the RISC processors. Thus, the assault from below, though it had suffered some initial setbacks, seemed once again poised to encroach on the commercial Unix market. And this time, the threat from below had an unexpected new component. The upstart Linux operating system burst into view and offered Unix reliability, not just for less money, but for free. Even though Microsoft's OS was still seen as having problems, Linux was regarded as highly reliable and absolutely adequate for low-end server tasks.
Further contributing to Sun's decline, was HP's and IBM's "re-commitment" to Unix. HP and IBM both recongized their enormous strategic blunder in cedeing the Unix market to Sun, and both companies sharply reversed course and began dedicating themselves to re-taking lost Unix marketshare. Both companies aggressively designed and promoted new machines to compete with Sun's. Both designed massive high-end SMP machines, HP's Superdome and IBM's S/80 (later, IBM's p690), to compete with Sun's E10000 and upcoming E15000. And both companies loudly proclaimed their sudden loyalty to Unix. Since both IBM and HP have more capital and more system-designing experience than Sun, both their high-end servers edged out Sun's in performance.
Thus, Sun was assaulted on three fronts. Sun's low-end marketshare was eroded by Linux and Windows 2000 running on commodity servers. Sun's high-end marketshare was eroded by new, more powerful servers from the freshly re-committed HP and IBM. And Sun's principal customers, "dot coms" and technology startups, disappeared as the tech bubble popped. These three factors were disastrous for Sun. Sun's sales and earnings plumetted; Sun's stock lost over 90% of its value; and Sun retreated back to the same position it occupied in the early 1990s, as a middling maker of Unix machines in a crowded market.
Sun's recent difficulties have led some to predict that Sun will fade into oblivion, going the same route as Digital Equipment. This danger may be exaggerated. Sun is financially well-managed and isn't in any kind of short-term crisis. But Sun faces serious long-term problems.
One of the most significant problems is the growing gap in performance between Sun's UltraSparc line of processors, and the processors offered by Intel or IBM. For a long time Sun's UltraSparc processor line had marginally trailed its competition in performance, but with the UltraSparc-III, introduced in 2001, the gap in performance considerably widened. This was due to two factors. First, Sun's CPU development team is less effective than the advanced and well-funded teams of Intel or IBM. Second, the process technology used to manufacture Sun's CPUs (by TI) is older than that found in the extremely expensive cutting-edge fabs of IBM and Intel. These two factors combined led to the UltraSparc-III CPU that is in-order, has longer pipelines, has fewer units, and runs at a lower clock rate than its competitors. The poor performance of the UltraSparc-III led some industry watchers to speculate that Sun is now incapable of designing and manufacturing high-end CPUs that are competitive with those from IBM or Intel. Since CPU design and manufacture are becoming even more capital-intensive every year, and Sun has far less capital than either Intel or IBM (and Sun's fab partner is willing to devote less of it), it's presumed that Sun will be unable to narrow this performance gap in the future.
Sun's response to this threat is to use an entirely new approach to CPU design, called "throughput computing." Sun claims that throughput computing will represent a revolution in CPU design and will allow chips that are 30x faster than those of today. The idea behind "throughput computing" is to replace the traditional massive, complicated processor core with a large number (>8) of small, extremely simple processor cores, each of which is capable of executing its own thread. Thus, a "throughput" CPU is actually a horde of tiny processors acting in tandem. The advantage of this approach is that it avoids the rapidly diminishing returns of expending exponentially more processor logic to extract smaller and smaller amounts of per-thread parallelism. Of course, there's a downside: single-thread performance suffers. But the tradeoff of single-thread performance for total instruction throughput is highly desirable for some applications. To most server applications, single-thread performance is unimportant, and total instruction throughput is all that matters. For these applications, Sun's new "throughput" CPUs might offer far better performance for a comparable die size and manufacturing process.
The difficulty is that Sun is more than one full process generation behind in manufacturing process, and it's unlikely that Sun's fab partner, Texas Instruments, is going to invest in new fabrication equipment on Sun's behalf. It's unknown whether Sun's "throughput computing" approach will offer significant enough performance benefits to compensate for older manufacturing techniques. Barring revolutionary advances in performance from the throughput apporach, the most Sun could hope for is to equal the performance of its competitors on server applications. Even if Sun's "throughput computing" proves to be a superior approach †for processor design, this advantage will likely be short-lived since there's nothing preventing IBM or Intel from employing the same ideas (Sun does not own the concept of SMT).
WILL THE SUN RISE AGAIN?
Thus, we come to the final question: Will Sun rise again? The three factors underlying Sun's prevoius ascent are permanently gone. The tech economy won't likely boom again like it did in the late 1990s; Sun's competitors won't likely try to kill off their own platforms; and Intel and Microsoft won't likely release entirely novel, untested versions of NT or Itanium. So, a repeat of the ascent similar to the one in the late 1990s is unlikely.
Furthermore, it's unlikely that Sun's low-end boxes will be able to compete on price/performance with the offerings from research-lean organizations like Dell. And it's unlikely that Sun's high-end boxes will be able to compete with offerings from massive, well-funded, technologically sophisticated companies like IBM and HP.
These facts have led some industry watchers to predict that in the long term, Sun is doomed. But if there's any lesson to be learned from Sun's history, it's this: extrapolating from current trends is not a good way of predicting the future. Sun is a master of pulling tricks out of its hat. Sun has been written off before, and it prevailed.