The reformation of Unix

Unix has been characterized as losing territory in the NOS market, but the venerable platform is preparing to recapture some territory lost to Windows NT, and it may be Intel’s IA-64 chip that lets it do that.

By Linda Stuart

It’s hard to think of Unix as the underdog in the network operating system market place, and yet time and time again that’s how it’s characterized.

NOS market study by industry research firms such as International Data Corp. repeatedly show Microsoft’s Windows NT taking more and more market share away from other NOS vendors, most notably the Unix suppliers, and when software vendors announce new server-based applications, more often than not the first platform supported is Windows NT, followed by staggered releases for different flavours of Unix, with support for NetWare usually announced as an afterthought.

What makes all of this hard to fathom is Unix continues to be the most powerful of server operating environments, the most reliable and the most secure. And yet, Microsoft has managed to get many network shops to switch to NT, by touting its NOS as being easier to use and more cost-effective than Unix.

Unix vendors are now looking to the 2000 release of Intel’s next-generation IA-64 microprocessor (code-named Merced) to help reverse this trend.

Commodity systems

One of the main reasons Windows NT is cheaper to implement than Unix environments is that the base Intel systems it runs on are cheaper than Unix vendors’ proprietary hardware platforms.

With the 64-bit IA-64 processor, Unix vendors will be able to run on the same commoditized hardware platform as Windows NT. As such, the base price advantage that NT enjoys today will be a thing of the past.

At the same time, Unix vendors such as Hewlett-Packard, Sun Microsystems, and the Monterey Project pairing of IBM and SCO are in the midst of porting their existing 64-bit Unix platforms to IA-64, giving them a significant headstart on Microsoft’s 64-bit NT plans. Microsoft has said it will ship a 64-bit version of NT when IA-64 becomes available next year, but by then Unix vendors will be delivering their second-generation — and in some cases, third-generation — 64-bit operating systems, said Tony Iams, senior analyst at D.H. Brown Associates Inc. in Port Chester, N.Y.

“The reason IA-64 is so important is that it could create a shift in the server industry,” Iams said. “When IA-64 comes to fruition…you’re going to potentially get leading-edge performance at commodity price points.”

Iams explained that despite Microsoft’s current dominant position on the Intel server platform, the NOS vendor will probably lose some ground to Unix once IA-64 systems ship.

“NT’s functions have not been progressing at the rate that Microsoft has expected, nor that users have expected,” Iams said. “So there’s a little bit of discontinuity here, in the sense that you will have Intel delivering high-end server horsepower, but the traditional operating system that was being used on Intel has not kept up in terms of its capabilities — in terms of 64 bits, in terms of reliability, in terms of scalability.

“This has created an opening for Unix to fill in that gap,” Iams said.

Commodity downside

The downside to commodity server hardware for Unix vendors — apart from shrinking profit margins — is it becomes harder for Unix suppliers to differentiate their server environments in the market place.

In the past, each Unix vendor’s respective proprietary hardware platform was a major factor in what differentiated one Unix environment from another.

On the other hand, the upside of having a standard hardware platform from Unix customers’ perspective — apart from reduced server expenditures — is that it now becomes easier to assess the actual fundamental differences and advantages of the various Unix operating systems.

For example, if a customer’s main concern is the symmetric multiprocessing (SMP) scalability of the server environment, Sun Solaris’ current support for 64-way SMP may tip the balance in its favour. Hewlett-Packard HP-UX currently supports 32-way SMP servers and IBM AIX is just now achieving 24-way SMP capability.

According to Iams, other key Unix NOS differentiators include fault-tolerance capabilities, such as the ability to adapt to failure of hardware components, or resource management, such as being able to take a large SMP system and partition it so that different applications utilize different processors on a consistent basis.

“Even though [the various Unix flavours] will be running on the same IA-64 hardware, the degree to which they allow that hardware to be exploited by commercial users is going to vary by operating system,” Iams explained.

“So there is definitely room for differentiation there, and these suppliers know that.”

Another plus for customers is that because next-generation Unix NOSs won’t be tied to their vendors’ respective proprietary hardware, users will be able to pick and choose the best combination of Intel hardware and Unix NOS for their needs. Whereas before a customer with a substantial investment in HP server equipment felt obligated to remain an HP-UX user, the commoditized IA-64 server platform may allow customers greater freedom to migrate to another Unix operating system.

“In theory, it is going to be easier to make that transition,” Iams said. “In the past, the barriers were much greater, because you weren’t just moving your operating system, you were moving your hardware, too. You had to switch platforms, and that obviously was a much bigger decision.”

Migration issues

Each Unix vendor porting its NOS to the IA-64 platform will have its own set of competitive advantages and disadvantages going forward.

In the case of HP, its existing HP-UX customers will have the easiest time migrating to the new 64-bit Intel hardware because the IA-64 processor, jointly developed by Intel and HP, will be binary compatible with HP’s current PA-RISC microprocessor architecture.

“You can move your HP-UX applications, your PA-RISC applications, and run them unmodified, uncompiled, move them right over to the Merced (IA-64) platform and they’ll keep on running,” Iams said. “That is a significant advantage for HP.”

In contrast, users of IBM’s 64-bit AIX will have source-level compatibility with IA-64, but they will need to recompile existing applications on IA-64 to get most of the microprocessor’s performance benefits. SCO UnixWare users will also be able to run current applications on IA-64, but only after doing a 32-bit recompilation on IA-64 or running the applications in emulation mode.

However, application compatibility between IA-64 and PA-RISC comes at a price for HP. Of all the Unix operating systems that will run on IA-64, HP-UX is the only one that will continue to run in “big-endian” mode. Big-endian refers to the format for storing or transmitting binary data in which the most significant byte is read first. The “little-endian” byte-ordering format reads the least significant byte first.

“HP could write HP-UX so that it would be little-endian, but then it wouldn’t be able to run the (existing) big-endian applications,” Iams explained. By offering its current HP-UX customers forward-compatibility with IA-64, HP is alienating those users from all of the systems and applications that will run in little-endian mode, he said.

This will become a major disadvantage for HP if and when the Unix vendor community comes to an agreement about a common application binary interface (ABI) that would allow Unix applications to run across different Unix platforms.

“It’s hard to imagine how HP could participate in that, given they will be big-endian and the ABI will be little-endian,” Iams said.

“I can definitely envision the ability to move Solaris applications onto Monterey and vice versa, and Linux applications moving between both, and just mixing and matching applications from any of those platforms. But it’s quite likely that HP-UX will be shut out of that because of this (big-endian/little-endian) divergence.”

In conjunction with that, HP faces another hurdle in that it hasn’t done a good job in the past of securing OEM partners, Iams said.

“Whereas you see Sun, IBM and SCO diligently going around trying to rally up key suppliers to be on their team,” he said.

In fact, Monterey Project partners IBM and SCO probably have the biggest advantage in this area over other Unix vendors because SCO in particular has been very successful in garnering support for its platform, Iams said.

“What really gave the Monterey agreement some teeth was the fact that IBM got together with SCO which has an exceptionally strong relationship with the channel, partners and OEMs, and the developers,” Iams explained.

“[SCO is] probably second to Microsoft in terms of securing Intel-based application support. Sun can probably compete (in the application market), but that’s all on the Sparc platform right now.”

One challenge for all of the vendors porting Unix to IA-64 will be ensuring that applications are ready to go when systems based on the microprocessor ship. In fact, some industry analysts are predicting that the compilers, debuggers and software tools needed to optimize applications for IA-64 won’t even be ready until the year 2001 timeframe when the second-generation IA-64 chip, dubbed McKinley, is scheduled for release.

“In many cases, users are just going to wait for packaged applications to become available on IA-64 before moving to it,” said Jonathan Eunice, an analyst at Illuminata Inc. in Nashua, N.H.

However, Eunice doesn’t think migrating applications to IA-64 will be too arduous a task.

“For most users, it’s going to be a bit of a hassle but not a major disruption,” Eunice said.

D.H. Brown’s Iams added: “I can’t imagine that any application is going to break (when it’s moved from a current platform to IA-64).”

In the wings

While most of the attention surrounding IA-64 is currently focused on what it will mean for Unix vendors and Microsoft, there is another operating system looking to capitalize on Intel’s next-generation commodity hardware.

At LinuxWorld in August, Intel chairman Andy Grove announced that the kernel of a version of Linux specifically for the IA-64 architecture would be released to the open-source community next year. In addition, Grove said Intel has Linux running on IA-64 simulators in its labs.

Many in the industry are wondering if the prospect of a 64-bit Linux running on a 64-bit Intel microprocessor will create the biggest impact on the NOS market once IA-64 systems ship.

In fact, according to Iams, some of the Unix vendors are playing both sides of the field by participating in the recently created Trillian Project, a collaborative effort to port Linux to IA-64. The Trillian Project is funded by Intel and includes participation by Linux developer V.A. Linux, HP, IBM and SGI.

“You’re looking at HP and IBM hedging their bets by also helping with the Linux port, in addition to their own efforts to move their proprietary Unixes onto Merced, with HP-UX and Monterey, respectively,” Iams said.

“Linux is potentially a threat because it commoditizes not only the hardware by also the software,” he said.

However, Jean S. Bozman, research director for International Data Corp. in Mountain View, Calif., said she believes Unix vendors shouldn’t feel threatened by Linux at this stage of the game.

“If you look at commercial Unix, typically it includes things that certainly aren’t in Linux today,” Bozman said. “It took Unix from the mid-’80s to the early ’90s to get all of the appropriate security and availability features, and more recently to scale up to support higher processors.

“Essentially functionality like a journalling file system and a logical volume manager would be a couple of examples of things that really need to be in Linux in a commercial era, if you will,” Bozman said.

“There’s no reason they can’t be built into Linux — the only thing it would take is time and money.”