Sun turns its rays on big iron
But can it compete with IBM in mainframe territory?
Sun has taken on the mainframe world with its Ultra Enterprise 10000. What does this market need that Sun can provide? What elements does Sun lack? Perhaps it's all a matter of mainframe coexistence, not mainframe replacement. We examine where Sun's strategy is headed. (3,300 words)
Sun's release of the Ultra Enterprise, or Starfire, series of enterprise servers marked a significant step upward for Sun. For the first time, Sun had computers that would compete in the high end of the market.
"We define high-end servers as costing $1 million or more," says Steve Josselyn, research director for International Data Corp., a Framingham, MA, market research company. "Sun will be entering that market this year with the Ultra Enterprise 10000 series. Now they have the ability to provide a growth path for customers whose business will grow over time."
This isn't exactly a new direction for Sun, but it is moving upscale fast. Josselyn points out that Sun's enterprise servers only crossed the barrier into what IDC considers the midrange ($100,000 to $1 million) a couple of years ago. Until then Sun had been mostly a workstation and departmental server company by IDC's standards.
With the E10000 server Sun has the raw computing horsepower to play in the mainframe world and do it at a significantly lower cost per MIPS than a conventional mainframe -- and it has some of the management and reliability features corporate data centers (the glass houses) want.
"What the E10000 did was give Sun the entre that opened the door [to the data center], says Paul McGuckin, vice-president and research director in the server group of the Gartner Group, a Stamford, CT, research firm. "It appears to have a feature set that is appealing to data center managers. That combined with Sun's aggressiveness up until about three months ago has allowed them to compete."
To understand what's happening, and what's likely to happen, it's important to grasp one fundamental point. No one expects Sun to rout the mainframes the way Sun and its fellow Unix vendors did the traditional minicomputer. What a lot of people do expect is that with the E10000 and its follow-ons Sun will be able to sell systems able to do jobs traditionally done by a mainframe. IBM's S/390 mainframe business isn't going to collapse under the onslaught, but Sun may very well carve out a comfortable and growing part of what has traditionally been the mainframe's domain.
"Data centers are no longer homogenous places," McGuckin says. "They haven't been for a number of years. Hewlett Packard's been in the data center. IBM [with its the RS/6000] has been in the data center, but until now Sun has had a great difficulty finding [an entrance]."
The obvious question is, "Why?" Why put the effort into developing a specialized line of products that can compete in what is, after all, a small segment of the market -- the high end. A segment, moreover, protected by multiple barriers beyond simple technology?
The short-form answer is that the high end makes excellent sense in terms of Sun's overall strategy. As well as being the direction of the technology and many of Sun's customers, it also fits with Sun's perception of itself and where the world is going.
The longer answer is somewhat more complicated. One very visible part of it is the growing challenge from Microsoft in the departmental and small enterprise server market. As Sun has pushed into what used to be minicomputer territory, Microsoft is pushing off the desktops and into the world of departmental and small enterprise servers that used to be almost entirely the domain of Sun and its competitors in the Unix market.
The mainframe market represents a place where Microsoft, Sun's arch-rival, can't go. Microsoft's recent Scalability Day only underlined how far Microsoft NT is from being a truly scalable operating system. At present NT isn't as scalable as Unix, and it has a long way to go before it can play the mainframe. With Solaris, Sun has an operating system that is much closer to mainframe standards and with the Starfire architecture it has a platform that comes closer as well.
A less visible, but probably more important, part of the answer is the needs of Sun's traditional customers. With the growth of the Internet, data warehousing, data mining, and other operations, the those customers need platforms that can handle the load. Sun customers would rather not go through the pain and expense to convert their Sun applications and platforms to mainframes. To keep those customers, Sun needs platforms that can handle extremely large databases and heavy transaction loads and do it with mainframe-equivalent reliability, availability, and maintainability.
Technology plays a role as well. Advances in processors, such as Sun's UltraSPARC, make it possible to build machines with mainframe-class performance without traditional mainframe costs, complexity, and environmental requirements.
Finally, there is the Sun mantra that "the network is the computer." If that's true you need some awfully large systems somewhere on the network to handle the sort of very large databases and such that support the network. Mainframe-class machines play into this need and give Sun the final piece it takes to reach all the way from the desktop to the highest levels of the enterprise.
The E10000 is considered by analysts to be a close-to-mainframe system. But close only counts in hand grenades and horseshoes. Starfire goes a long way toward meeting the hardware requirements of mainframe type applications, but the software isn't quite there yet. Over the next year Sun will probably roll out most of remaining pieces.
One of the missing pieces is a 64-bit version of Solaris. According to Sun's road map, that is coming over the next few months as Sun phases in 64-bit features in new releases of Solaris. Sun has chosen to introduce 64-bit Solaris gradually because it wants to avoid the kind of compatibility shock it put its customers through when it went from SunOS to Solaris. To ease compatibility problems for existing customers, Sun decided to phase in 64-bit Solaris over a period of a couple of years, with each succeeding version having more 64-bit features. This made the existing customer base happy, but it meant that Sun is a little behind companies like HP that went directly to 64-bit operating systems.
A 64-bit operating system, however, is only the beginning of what's needed. One of the other critical requirements is mainframe-level system administration -- a significant step up from midrange servers.
"If you go into a data center you will see all these consoles, sometimes a whole wall of them," says Jean S. Bozman, an IDC analyst who tracks Solaris. "People are looking at how jobs are being processed, getting feedback on processor utilization level, seeing if the workload has to be rebalanced. In terms of system administration, those are some of the areas Sun is trying to enhance even further."
It isn't that Sun doesn't have most of these management features available, it is that it doesn't have them to the degree mainframe users expect. Here again, Sun is working hard to provide mainframe-class system administration.
The standards of reliability and maintainability in big-iron country are far higher than in enterprise-server land, and security requirements are much tighter as well. This requires a company like Sun to meet a whole raft of new challenges if it wants to move into the mainframe world. Sun's Solstice HA (high-availability) software is aimed in this direction. However, the product is still new and has some kinks to be worked out. As a result, some new installations are combining Ultra Enterprise servers with third-party HA software from vendors like Qualix Group.
One of the key elements in mainframe-type operations is reliability. Downtime is hideously expensive. According to one study by Contingency Research Planning, the financial impact of a brokerage operations system going down is between $5.6 and $7.3 million an hour. The cost of a credit card or sales authorization center going down is between $2.2 and $3.1 million an hour. A number of other businesses reported downtime costs of well over $10,000 an hour. In general, mainframes promise downtimes, including hardware replacements and software upgrades, of around five minutes a year.
Getting this kind of reliability and maintainability means, among other things, providing a way to hot swap everything from disk drives to processor boards to software while the system is still running. Sun now has some of this, but not as much as a mainframe -- or even some of its enterprise server rivals. "HP and IBM high availability products allow you to support rolling operating system upgrades," McGuckin says. "In other words if you have a four-member cluster, you can take one member down, install a new level of the operating system and bring it back into the cluster. Then you can work your way around the cluster doing this. Sun can't do this and it doesn't even show up on Sun's road map until a couple of years out."
Sun has been emphasizing reliability, maintainability, and serviceability in discussing the E10000 system. Significantly, when Sun announced the E10000 multinode system recently it stressed its use for increased reliability rather than increased capacity. That makes sense to John Young, vice president of enterprise systems planning for The Clipper Group, a Wellesley, MA market analysis firm. "There are very few enterprises out there filling a 64-processor Sun 10000," he notes wryly. He says, however, that the failover capabilities of the a E10000 in multinode configuration are the sort of thing data centers look for in this area of the market.
Another factor is user conservatism. Given the demands they put on systems, high-end mainframe users are very cautious about switching to new platforms. Not only are the costs of conversion enormous, even a seemingly minor glitch can have major repercussions.
The tendency in mainframe shops is to keep the big computers and offload new kinds of applications, such as browser-based GUIs to specialized servers. Essentially, the mainframe becomes a very big, very fast database server. In this kind of data center environment, mixing mainframes and Unix servers, the ability to share data more closely between a mainframe and a Unix system becomes increasingly important -- and not easy at the performance levels needed.
"The traditional strategy has been to ship that data either by sneakernet (carrying magnetic tapes between machines) or by networking, which is both slow and impacts the workload that really requires the network," Young says. "It's been possible to do it, but it's been a tedious, manpower- and resource-heavy kind of operation. Now, as data centers become more duplexed there are efficiencies to be had by finding some other way of data transfer."
According to Young, this is one of the major factors in Sun's decision to buy the storage business of Encore Computer Corp. (Ft. Lauderdale, FL). "Encore's solution, rather than making multiple copies of the data, was to find a way to keep one copy of the data and be able to share it dynamically between the mainframe and the open systems. That's the technology Sun is trying to buy through the acquisition of Encore's Infinity product line. When Sun can integrate that technology, it will give them a significant entre into the data center."
The death of the mainframe left a very lively corpse
This doesn't mean that mainframe purchases are standing still. In fact, sales (measured in mainframe MIPS computing power) have surged in the last two or three years thanks to the need for additional power in the glass house and falling mainframe prices.
According to IDC, mainframe MIPS shipments grew by nearly two-thirds from 1995 to 1996. Last year companies added 535,000 mainframe MIPS of computing power, the highest growth rate in nearly 15 years, IDC says.
"The media talk about death of the mainframe turned out to be both premature and overstated," Young says. "There hasn't been a wholesale retreat from the mainframe and a wholesale return to them."
Indeed, Young says, a lot of the talk was based on a misperception. "If a mainframe was displaced completely it was a small one that was two or three generations old. Or it was a case where one or two applications were taken off the mainframe and put on a Unix platform."
The mainframe's future wasn't always this rosy. According to IBM's numbers, its sales of mainframe (S/390) MIPS peaked in 1991, declined gradually for several years and then took off like a rocket in 1994. A large part of the reason, IBM says, is new applications on the mainframe. Specifically, the company says that something over a third of the new S/390 MIPS are going into new applications and that mainframes are selected as the platform of choice for new applications almost half the time in shops with S/390 mainframes.
All of which probably means less than it seems, especially in terms of where new applications are going. However, it is undeniable that mainframe shops are adding more mainframe power and that a lot of that power is going to new applications.
One of the big reasons mainframes haven't died is that they have adapted. In the late '80s and early '90s, IBM and the other mainframe makers came to understand that what their customers wanted wasn't so much bigger, faster mainframes as it was cheaper mainframes. While the top of the line did get bigger and faster, almost all classes of mainframes got a lot cheaper. The mainframe makers did this by adopting the strategies and technologies of their competitors -- like Sun.
Among other things, IBM now has a line of mainframes built with CMOS chips, like workstations and PCS. CMOS technology is slower than the older ECL (Emitter Coupled Logic) designs, but CMOS machines don't need the elaborate water cooling systems of conventional ECL big iron.
One result has been that mainframe prices per MIPS have plummeted at about 30 percent a year. Where a mainframe MIPS on IBM's System/390 line cost about $23,000 in 1995, it is less than $14,000 today. (Another result is that the mainframe business has been growing solidly since 1994, but mainframe profit growth has stayed at about 5 percent or less.)
Ironically, another of IBM's responses to the challenge from companies like Sun has been Unix. The latest version of the company's MVS operating system, MVS OS/390 supports Unix APIs. This makes it easier to migrate Unix applications to the mainframe, if a company desires.
Another critical factor is the cost of converting data from mainframe to Unix systems. IBM likes to say that 70 percent of all the world's data is stored on mainframes, and most of those are IBM mainframes. Certainly there is a tremendous amount of information on existing mainframe systems.
"Companies have made a tremendous investment in mainframes over the last 30 years," Bozman says. "The fact that they work and the fact that a lot of companies have already paid for their mainframes mean they aren't in any rush to replace them."
What it comes down to is that direct replacement of mainframes is a small market, and it will probably stay small for years. Replacing a mainframe with a Sun enterprise server means converting software, retraining staff, and other expensive measures. Most of Sun's potential market in the mainframe space is in additions rather than replacement. These are new systems that are taking over a new or greatly expanded job that would traditionally be done by a mainframe. This is especially true of Unix users who need to support much larger applications but want to put off moving to a mainframe for as long as they possibly can.
One of the other factors is that the role of the mainframe is changing as well. "Where the mainframe used to do everything, now the mainframe is more of a database server. At the same time it's network central. Often the mainframe controls a lot of features of the network," Bozman says.
Comments Young: "You're finding more differentiation by application type than by technical capabilities [between Unix systems and mainframes]. The decision on which to use is being made more by enterprises based on where they want to run specific types of applications. Some applications can be more efficiently run on Unix and some on the traditional mainframe."
IDC's Josselyn generally agrees. "There are certain types of work loads that the S/390 base has typically not been a good player in and where Unix environments have done a lot better. That includes data warehousing and decision support. With those types of workloads they can go in and implement a new application on a bigger Unix server."
On the other hand there are some applications that are staying on the mainframe. As an example Josselyn cites high-end transaction processing. "I expect that will be the case for some time," he says. "They [the mainframe makers] have been geared toward the big transaction environment for many years. They understand that business. Because the system management capabilities are already in place and because the expertise already exists in those organizations, no one is going to take it away now.
However, Josselyn points out that "down at the low end of the transaction environment Unix plays very well and is very successful."
What is perhaps less clear is what happens as organizations now using small- to medium-sized Unix systems for transaction processors grow and need high-end transaction processors. Will they replace their existing systems with mainframes as they had to do in the past? McGuckin points out there are already cases where a Unix server competes head to head against a mainframe in the data center, such as Enterprise Resource Planning (ERP) software like R/3 from SAP.
In fact the division between mainframes in their new roles as super servers and ever-more-powerful Unix systems is starting to blur the traditional lines between the two. "You're going to see a convergence," Bozman predicts, between Unix systems and mainframes.
Of course Sun isn't the only company eyeing the mainframe market space. Hewlett Packard and IBM's RS/6000 division, among others, are introducing systems and software designed to compete in that arena. At this point, Young says, there is no clear front runner, although HP and Sun are very close.
Looks good, but...
Only time will tell if Sun's push into the high end will be successful. So far the signs are favorable and most industry watchers seem to be impressed. However, this is still a very new market for Sun, and the customers are both savvy and disinclined to change for change's sake.
Sun has well-regarded hardware and software, a good reputation in other areas, and a convincing story to tell data center operators and CIOs. One indicator will be how well the E10000 does in this, its first year on the market. Says Josselyn: "I think we need to see what the results of 1997 are for the high end of Sun's product line. Then it will be easier to judge how successful they will be up there."
About the author
Rick Cook divides his time between writing about the Web, computers, and high technology, and novels. His most recent stories for SunWorld are "Running Windows 95 under Solaris -- Is it smooth sailing now?" (May 1997) and "Developers at a crossroads: How will Apple's incorporation of Next impact the future of NextStep for Solaris?" (March 1997). Reach Rick at firstname.lastname@example.org.
If you have technical problems with this magazine, contact email@example.com