Firing up IBM's Unix business

Because there's money to be made, according to Ross Mauri, who took over as general manager of IBM's System p business in January. Unix servers are powerful, often expensive systems used at the heart of customer computing operations. They're on the frontlines of IBM's battle with Hewlett-Packard and Sun Microsystems.

Mauri, 47, is a technologist who spent years in Big Blue's mainframe group and who led overall server development before taking over the Unix group. It's a nice fit: Brawny Unix servers are a good home for the reliability, efficiency and performance features originally developed for mainframes, and now being spread to other IBM server families.

Mauri took over the Unix server group just as IBM completed a years-long climb to the top of the Unix server market in 2005. But the company's Unix share slipped in early 2006.

The Unix chief is confident he can reawaken customers' appetite for spending on high-end IBM systems by bringing the Power5+ processor to the flagship p5-590 and p5-595 servers. He discussed IBM's views on technology and competition with CNET News.com.

Q: The overall Unix server market is pretty much flat or shrinking. Why devote a lot of attention to it?
Mauri: It's been flat or slightly declining for a decade now, but it's a big, important market. Given the amount of work and applications across many, many different industry segments, we see it as a great land of opportunity. We're going to continue to attack it vigorously, so it's worth the investment.

Power6 Systems are scheduled to appear next year, so I presume the Power6 chips themselves are probably pretty well along. What state are they in?
Mauri: So far, it looks pretty good. They are up and running in our lab, running at (expected) frequency. We're very pleased with what we see so far, with good yield (the percentage of manufactured processors that are usable). There is still a lot of system-level testing that needs to go on, with the complexity of today's systems.

IBM has said Power6 will run at least at a 4GHz clock speed. Is that still the plan?
Mauri: Yeah, 4 to 5GHz for sure.

Have you changed your designs or your direction with the current ruckus about performance per watt?
Mauri: Energy consumption clearly is on the minds of our customers in the marketplace right now--rightfully so. We've got some pretty interesting things that we've been in working on to have more power-efficient management of the overall systems architecture. You'll see us continue to be sure that we've got the best wattage count for the workload done. And we'll start to do some interesting things with how we manage the power overall, so customers can decide about how they want to manage the overall power usage in their data center.

Something that would monitor power consumption and performance and give customers a dial so they can ratchet power down when necessary?
Mauri: Yeah, absolutely, including tying it to some policy that they might set--whether it's time of day or time of month, if there was particular heavy demand that they knew of, or just reacting in real time.

Sun managed to turn things around a little bit recently and gain some market share at IBM's expense, and I noted also during their second quarter of 2006, Sun sold $100 million worth of their UltraSparc T1 "Niagara"-based servers. Do you see Sun as a resurgent threat?
Mauri: After 10 straight quarters of decline, I guess their compare (comparison with year-earlier results) has got easy enough that they were going to bounce back a little bit. I have seen what they have done with their Niagara Servers and their Galaxy Servers, but I think they are still struggling. They did make some of those Niagara sales in their installed base.

So their market share gain was just a blip, and they're still destined for doom?
Mauri: I would like to think so. This market ebbs and flows. If you look at our performance for the last five years, we've had a slight slowdown when we're doing product refreshes, like we were this year with the high-end transitioning to Power5+. In those times, Sun has done a little bit better. I think that you're going to see us come on strong with this new set of products.

Intel has announced Montecito, its latest Itanium processor, and the chip will be showing up in systems in the next month or two. What's your assessment of how Intel has been doing with that processor and how Hewlett-Packard has been doing with its Itanium-based Integrity line?
Mauri: We're doing pretty well at beating them head-to-head. With their refreshed products that will come in the September or October time frame, I'm sure they will get some PR out of it. But frankly, I think that our technology and our performance leadership are going to prevail here. I think that Itanium is a bit of an anchor for HP, with their PA-RISC base and their Alpha base (two processor lines that are being discontinued) of customers being forced to transition.

With Niagara, Sun has aggressively pushed performance-per-watt and multicore processors that run lots of instruction threads. IBM has chosen with Power to use many fewer cores and many fewer threads. Do you think that Sun's strategy has any merit or are you confident that heavy cores are the way to go?
Mauri: Multicore is something that we pioneered, and you'll continue to see us make great strides in that area. We've also had multithreaded systems for a long time. I'm really comfortable that we've got a winning strategy. I don't expect some big switch from IBM's point of view surrounding this.

Power6 is another dual-core processor, correct?
Mauri: Yes, Power6 will be dual-core.

And do you think you will go quad-core with Power7?
Mauri: We're looking at higher levels of cores for chips, as we demonstrated in the Cell processor. That's clearly demonstrating multiple cores and being able to run a high amount of workload.

One interesting change you made with the Power-based systems in recent years was the addition of Linux as an operating system. What are your long-term ambitions for Linux--versus your version of Unix, AIX--on Power?
Mauri: That's good question. People have been asking us that for since 2000, when we started really talking about our Linux strategy.

We want Linux to have a great home on the Power architecture. We will continue to contribute to the open-source development of Linux, as well as all of the packages that surround Linux, to ensure it performs well on Power. But as we have said, AIX is our flagship on System p. We've invested a lot in it. Nothing scales as well as AIX. Nothing is robust as AIX. It leverages all of our virtualization technologies and, quite frankly, most of our customers worldwide have bet their business on it. We're going to continue to invest in AIX and continue to enhance it.

So is Linux a red-headed stepchild? If AIX is the flagship and the priority, why should customers mess with Linux on Power?
Mauri: Linux on Power can be used in conjunction with AIX, running virtually next to one another on a single box. So for consolidation, you (can) bring in some Linux workload.

Sometimes, customers are consolidating and taking out Sun Solaris, and they decide to convert those Solaris workloads over to Linux. We're also seeing Linux very healthy and happy running on Power in some standalone applications where the customers decided that Linux was good enough--for instance, some of the things we've been doing recently with Sybase and their IQ product.

We're going to treat them both properly and support them both. There doesn't have to be an absolute, one or the other. The customers that want both, and we see good return on investing in both.

Do you ever consider open-sourcing AIX the way Sun is open-sourcing Solaris?
Mauri: No, we're not. I think that OpenSolaris is a little bit of a game Sun is playing to try to get good PR. But I don't think it's in the spirit of true open source.

We have been very happy to get directly involved and contribute to Linux and Apache and the Eclipse Foundation. We're not going to open-source AIX. It's best run on the current model, where we have the expertise. We enhance it. We work closely with our customers to listen to their requirements. But in the end, it's best that we control that source code.