Linux on Power

By eweek  |  Posted 2004-03-01 Print this article Print

Some industry observers see IMs work with Linux on the Power platform as a gamble of sorts that, while it makes sense, you risk becoming distracted away from other aspects of your systems business. Can you talk about the importance of Linux on Power for IBM? If you look at whats happened in the server market over the past couple of years, most of the revenue decline came from the RISC-Unix platform, and most of that came from four-way and below as they were cannibalized, or substituted by Linux on Intel or Linux on virtual partitions. Most of that came from Sun and HP, because they have the strongest positions there, so cannibalization for us is more of an opportunity than a threat because AIX didnt have the position there that Solaris or HP-UX [did]. So we think that Linux gives us the opportunity to get more work onto our platforms.
When we looked at BladeCenter, something on the order of 65 or 70 percent of the first 50,000 that were shipped ran Linux, which is considerably different from what Intel servers would have been in general. So our reasoning is, if it runs Linux, there are some workloads that run on Linux that would do better on Power than they would on other kinds of things and so it gave us a pretty good opportunity. I dont think that Linux represents a cannibalization threat for our Power systems. What it represents is a workload opportunity, and thats the way were approaching it.
I think here the traditional logic that would say that its difficult to get software vendors to support a new operating system on a platform doesnt apply quite as much. First, its simpler to move from Linux on one platform to Linux on another, just like we did it on the mainframes. The fact is that most of the applications that youre going to be trying to move arent traditional SAP [AG enterprise] application servers or PeopleSoft [Inc.] servers or things like that. Theyre infrastructure workloads—Web servers, file servers, cachers, firewalls, things like that—many of which had their origin in the Linux world. So when you look at it in terms of pockets of opportunity, weve convinced ourselves that there are segments where this would be attractive from a price/performance standpoint, and I dont think youll see it disrupting our current Power activities as much as complementing it. The last part of this is that an important element of Squadron is a common hypervisor that runs on all of these platforms—[pSeries, iSeries] and storage. That hypervisor, just like on mainframes today, allows you to run different kinds of operating systems—OS/400, Linux, AIX, storage subsystems. So, just like you saw on mainframes, people running multiple Linuxes to consolidate workload onto virtual servers, and just like weve seen surprising acceptance of this idea on our iSeries platform of inboarding Linux onto virtual servers, I think you will be seeing Linux workloads inboarded to pSeries systems with AIX, which is our whole goal here—get more work onto these systems. It is one of the pivotal things we have to do this year. How will IBM get Linux really moving forward on Power? There are three areas youd want to have capabilities. One is virtual servers on larger SMPs [symmetric multiprocessor systems], and I think our capabilities are pretty good there. The second is blade servers, and we have started limited shipments of those, and we will have them in volume in the second quarter, and the third is stand-alone one-, two- and four-way servers. Now weve had some of those in systems that were originally designed as AIX systems, the [p630] and things like the [p615]. In the first half of this year, youll see more high-volume stand-alone—that is not big SMP partitions and not blades. And I think were going to need to do all three of those things to be attractive to people. We have plans to work in each of those three. The other reason that the blade is interesting is because its the first time that we used the PowerPC 970, which is the same processor thats used by Apple [Computer Inc.] in the G5, and part of our strategy here was, if you are really going to be broadly successful in high-volume Power, youve got to have a different design and cost point. Prior to this, we were trying to make the one processor fit everything, from 64-ways down to blades, and that was kind of hard recognition for the technical team to get themselves to. But now that theyve gotten themselves to it, I think that youll see us accelerate the delivery of things that are optimized to the low end. One of the reasons we did better on the high end and didnt do so well on the low end was that our design point for these Power4 and Regatta systems, we consciously started to capture the high end. We didnt design for the one-, two- and four-way space, really, we accommodated them. Now that was good for us when Linux came in and cannibalized the whole thing, but if you say, "How are you going to take advantage of all this?", some of the things that really are very, very important at the high end that cost that maybe you dont want at the low end. So we made another important breakthrough in the middle of last year, which is … maybe we should use the other elements of the Power family [and] drive off of the high-volume products that we have. In the one-, two- and four-way space, we had products that were Power-based, in the [p615] and [p630], but in the middle of the year youll see built off these 970 products much better optimization into that spot, cost and performance optimization. Next page: Lukewarm embrace of Itanium and Opteron?


Submit a Comment

Loading Comments...
Manage your Newsletters: Login   Register My Newsletters

Rocket Fuel