Hello, this is James Governor here at Redmug TV at the IBM Z-Summit for 2008. Would you like to introduce yourself, John? My name is John Shedletsky, I work for IBM in the Software Group, and I'm the Vice President of Competitive Technology. Historically, IBM didn't always do the best of jobs at helping its customers to make the economic case to their businesses in order to actually keep with the mainframe. I've noticed a bit of a change in that. Is this something that you're partly responsible for, this fact-based ability to help the customer make their case? Is that your group that does that? Is it a coherent strategy, or is it just something that I'm going to... Well, you know, for some time, the hardware organization in IBM, which is referred to as STG, which produces the main hardware product, NZOS, really drove these discussions about TCO. And in the software group, I think Steve Mills came to realize that you really can't talk about TCO in isolation from the software. It was really a combined story. And that was one of the assignments we got when he asked me to start getting involved in this subject. I will say that I think we've been a little more transparent with our data than has been the case in the past. I've had customers tell me, you know, IBM's pricing policies on the mainframe have been rather opaque to me. And I think this, in some sense, it hurt us rather than helped us. Absolutely. Because in the absence of information, competitors come in and say things that may not necessarily be true, and there's no one to contradict it. So we decided we needed to be a little more transparent. Well, frankly, exactly to your point. Whilst, yes, you may have been creating barriers to entry for your competitors, you were also creating barriers to participation from new customers.
2:00What have you found out? What's the story? What are the economics of this mainframe? Well, it boils down to one main idea, and that is the idea of consolidation. It's linked to this notion of economy of scale. It's a fundamental principle in economics. Even Henry Ford discovered this. Well, he didn't discover it, but he exploited this concept maybe a century ago when he created the automobile assembly line. The principle there was to get all the resources in one place and to be able to produce a unit of production at the lowest possible cost per unit. And the way he did that was by increasing the utilization of all the resources, namely the people who were working on the assembly line, and minimizing wasted effort to get all the energy focused on producing that production unit. The same is true with computing. In computing, a unit of work might be a transaction, and the objective here is to produce a transaction at the lowest possible cost per unit of work. And the way you do that is to get higher utilization out of the resources that are involved. And the way you do that is to run multiple workloads, because of the nature of how a computer works, CPU, having to wait for disk operations, and so forth. The way you actually do that is to run multiple workloads at the same time, multi-programming, and to apply some level of intelligent workload management to it so you meet service level agreement. The mainframe was designed to do this from the beginning, whereas some of the other platforms, most notably Windows, has really been regarded as a single application environment until very recently. And in the Unix world, while you could run multiple processes, in practice, even Unix servers were more or less dedicated to a particular application. It's really been the mainframe alone that really made it an operational principle that its job in life was to run multiple workloads simultaneously and to achieve very high levels of utilization of the process.
4:08Let me ask you about that, because that's a real challenge. One of the things is that most of the benchmarks are based on single proc, or at least single machine, single workload. How do you do that? Because there aren't really benchmarks for mixed workloads. Some people do benchmarks, but benchmarks are inherently interesting. They'll give you insight into processor speed, for example, but they don't necessarily tell you what you want to know. Because in a benchmark, what you do is set up some application program, and then you drive it to 100% utilization. Absolutely. But this is the point, that on these dedicated resources, typically they don't operate at 100% utilization. Whereas a mainframe can run multiple workloads, achieving 100% utilization of its processes. So, you can't just compare head to head, you know, let's just run this application just as hard as we can on both systems and see what throughput we get. You have to more step back and say, what's the workload in the data center? It becomes more of a data center issue than a platform issue. So, for example, that's why IBM is one reason why IBM has been focused on this so-called new enterprise data center that we've been talking about. Because we realize that the issue here is getting the lowest cost per unit of work out of the data center, all of the workload that the data center does. And what is the best way for a client to go about doing that in their data center? That's a possibility. I would say we could review where we are right now. Right now, I think primarily the role of the mainframe is to manage the core data of the business and the transactions that execute against that data. That's been the traditional role, the traditional strength. That's what the mainframe has been optimized to do for many years now. And we still see that. The mainframe is used by 66 out of the top 67 financial service companies in the world to manage their transactions against their core data.
6:16That's its strength. And it can do that in a way that other systems can only approximate. So, I see that role continuing. Do you have any messages for the London Stock Exchange? I saw something in the paper about a crash there. And it followed shortly after another article I saw about a crash at the Malaysian Stock Exchange going down for nine hours. Okay. They were running HP non-stop. Both of them, in fact. And it stopped. Yeah. Not so non-stop. Not so non-stop. Not so non-stop. So, yeah. So, I saw some back and forth about the London Stock Exchange. I weren't sure what to blame for the outage, but they were running non-stop as well? Mm-hmm. Well, what they have, the core trading platform is still non-stop. They also have a messaging platform, which is Windows-based. I think what came back to hold Microsoft in that is they've done a brilliant job of PR-ing it that they run the exchange. So, of course, when the exchange goes down, everyone says it's Microsoft's fault. They were associated with it. Yeah. You rarely hear about events like that with businesses that are running a mainframe. We have examples of customers, I know, of a bank that has not had an unplanned outage in ten years. Ten years. It just runs. And the only way that you can achieve that kind of availability is a layered strategy that starts at the heart. You mean you don't have to reboot mainframes every day? No. It just runs. Just like it should. I mean, if you're running your core business and you have a global business, you need to be able to run that core business 24 hours a day. Mm-hmm. And this is one of the key values that the platform provides. Investment we make in availability would be hard for any other vendor to match. It starts at the design of the chip. There are things in the chip that are there to provide redundancy. What the earlier speaker today was talking about manage, what did you call it, oversight of the memory usage and oversight of the processor execution. These are error detection and error correction and access control mechanisms.
8:20And it goes from there. There's reliability and serviceability. If you do need a part anywhere around the world, most likely the system has already detected that and called IBM. And most likely that part is positioned somewhere close to the customer anywhere in the world. Mm-hmm. So we have global serviceability. And then going on from there, you look at things like the sysplex that enables clustering on a scale that no one else can really -- a scale and efficiency that no one else can match. So that even if you lose a system, if you do lose a system, the cluster continues to run. The software is designed within the cluster to be upgraded without bringing the cluster down. Many of the maintenance operations on things like DB2 can be done without bringing the database down. So you even eliminate planned out. There's a layered strategy that we've been working on for years with a lot of feedback from our customers. Mm-hmm. Just honing off that last little risk or that last little -- Continual improvement, basically. Continual improvement, so that these platforms will run reliably. To bring it back to software a little bit, one of the things that I know that many organizations don't fully understand is the cost benefits, to your point about software, and if we think about mainframe licensing, which was sometimes opaque and occasionally a blunt instrument. Mm-hmm. That there are some advantages of running the application service, say, and the database on the same box. By collapsing the tiers, there are some clear and substantive benefits, and a lot of people have not realized that. I refer to that as co-location. So take a typical case. You may have your database running on a mainframe, and people will often think they want to extend that -- they want to extend the channel to the Internet for that data, maybe a customer service channel or something.
10:23And they will typically maybe use WebSphere and think about running WebSphere on Intel or Unix -- Because it's cheaper. -- servers sitting in front of the mainframe. Well, we'll get back to the cheaper in a minute. But we've done some benchmarks on this, and we have seen that basically the closer you can run that channel extender, in this case a web application, to the source of the data, the better performance you're going to get. So you can move it from a separate platform onto the Z, you can run it in a separate LPAR on Z, separate partition, or you can run it even in the same partition. And the closer you can get, the better your performance will be, and we refer to that as the co-location. And it can be a dramatic improvement, one and a half, two times. But also, of course, in licensing, there is an advantage as well. Well, in licensing, that's a little different story because, for example, you have to consider what would I pay to run this on Z Linux, where I'm taking advantage of the IFL, running 100% on an IFL. And I'm only paying for the software on a per core basis, a per IFL processor basis, which is just almost the same as what you would pay to run it on an Intel dual core. So that's a very low price there on Z Linux. You can run it on ZOS in the same partition as DB2. Right. But then you would have to take advantage of a ZAP processor, and while you may get 85% offload to a ZAP processor, there's still that 15% that's going to generate NLC. So the cost equation is a little bit different, and customers that are motivated by one or the other will make different decisions. It's still kind of difficult to compete on the understanding of TCO against a Microsoft or a Solaris platform. So how can you get better at that, especially in terms of the language that you use?
12:23That opacity remains. IBM has this language, a self-contained sort of church language. Other people don't understand, and that is a barrier to entry and participation in customers. So you say we say the liturgy of LPARs and MIPS. That is correct. Yeah. We do do that. I've struggled with this question because I think there are fundamental differences. When we talk about the new enterprise data center, that's all, well, it's about a number of things, but cost is at the center of it. You could look at what Microsoft is doing. Microsoft has 15 data centers today around the world. So obviously they're running Wintel. They don't run on anything else but Windows on Intel. And I believe they have a hundred and -- Actually, you can run Windows on Solaris now, but -- Well, you could, but they don't. And I think they have a total of 150,000 servers right now supporting their operations. And they're trying to move into this software as a service business model to compete more with Google. And they're planning to build 24 more data centers, 500,000 square feet each. And they recently had a conference where they talked about the plans. Now, the thing that I find amazing about this is that in order -- They understand these issues of electrical consumption, cooling, floor space, and cost. Because their data centers are getting so massive. Even for Microsoft, the cost is massive. So they're trying to do something about it. And what they're doing is they're abandoning the HP Intel servers, which is what they run now. I believe it's ProLiance. And designing their own custom servers. And they're packing them in blade-type configurations into racks. And stuffing these racks into what effectively are shipping containers.
14:26I sit back and I think about that. So here's a company that predicated on the idea that the Wintel platform is the cheapest platform you can run on because it's commodity servers. And now they've gone down this road, like Steve was talking about this morning, where some of the other accruing costs start to accrue. And they've realized that they can't do that anymore. They have to build custom servers, just like Google has built custom servers. So this is where the Microsoft endgame takes you, to custom servers and blades packed in containers. The points I'll make about that: that nibbles at the edges of the cost equation. It doesn't address the most important elements of cost, which is labor and software. Now the ironic thing is, if all their software ran on Linux, let me talk about the alternative. The alternative is to consolidate. So where Microsoft sees their consolidation container being a shipping container, where you have to leave a little space in there for a person to walk in and deal with all those blades, our consolidation container is a ceramic module in a mainframe. With layers and layers of wiring, where the chips are really close to each other, and where we can power them and cool them in a very efficient way. It's back to this idea of consolidation as a far more effective vehicle for reducing cost, and all these other side benefits, like energy consumption, than simply just packing more blades into a tighter and tighter space. The second point I wanted to make: you could have a thousand blades in a shipping container, and you still couldn't do transaction processing against a single large database image. There is no technology in the world today that will let you do that. So you could say that Microsoft's solution for their data centers are really customized solutions for the kind of services they're contemplating providing. Maybe it's a different kind of workload to be fair.
16:26Different kind of workload. A final question would be, where do you see the mainframe in 10 years? I was having a discussion with someone about where do we see this idea of data centers in 10 years? The logical ultimate... Are you going to tell me there will be five? We're back to Thomas Watson. Two. Two. Two in the world. Just this so-called, this concept of cloud computing where you basically rent a server, is very popular in the Bay Area. My son works for a startup and they rent their servers. You don't even know where they are. Right. He just rents them out of the cloud. And, you know, the logical progression here is to consolidate everything, as long as you have security for isolation and so forth, into a few mega cloud centers around the world. Look at it from the customer's point of view. This is great. They don't have to worry about getting power delivered to their data centers. They don't have to worry about having to build a new data center. They don't have to worry about financing to finance all the equipment that they need. Well, as Nicholas Carr points out in the big switch, and, in fact, Jonathan Schwartz from Sun has made the point many times, you know, we used to have chief electricity officers. And manufacturing organizations would have their own electricity plants. That's not necessarily an ideal model. The flip side is, increasingly these days, we're seeing companies, you know, BT investing in wind farms, Google investing in renewables. Perhaps the chief electricity officer will come back. So, I'm not so sure we're going to see these five data centers, but it's an interesting vision, and it could be that, yes, that Thomas Watson was correct on that famous quote, you know, the world doesn't need more than five computers. If that's the vision of the future, a lot of consolidation across enterprise boundaries, because of the economics that that provides, then I see System Z being at the heart of those consolidated centers.
18:26Now, that means it'll have to be able to run more different kinds of workloads, probably different architectures. You could be running windows. That are tightly integrated the way that the processors are today, sharing at the ceramic level, not at the container level or at the network level. So, that was Red Bug TV and John Chemniewski. Thank you very, very much. You did. Thank you.