Q&A with Erich Clementi, zSeries General Manager

by Timothy Prickett Morgan

Erich Clementi, the general manager of IBM's zSeries mainframe product line within its Systems & Technology Group, is a hard-core mainframer, just like many of the other executives who have run the Enterprise Systems Division, S/390 Division, and zSeries Division before him. Having a broad and deep understanding of how the mainframe fits into past and current IT infrastructure--what some might still call data processing, if they were honest--as well as some wit is what makes Clementi stand out among his predecessors.

Clementi took some time out of his busy schedule recently to talk to me about the state of the mainframe business and where he sees IBM's big iron heading in the future as the company continues to do engineering and product development to meet the high-end data processing needs of mainframe shops.

Timothy Prickett Morgan: Tell us a little bit about yourself. What is your background at IBM and how do you get to be general manager of the zSeries business?

Erich Clementi: I am Northern Italian, and I've been with IBM for 21 years. This is the fifth country that I have worked in during the past ten years. I spent my first 15 years at IBM working in the banking and financial services sector in Italy, Germany, Switzerland, Austria, and Eastern Europe. I spent my first four years as a good, old-fashioned systems engineer. So I programmed mainframes and I installed software, and after that, I moved into other areas. But when you are working in the financial sector, you are always in the vicinity of a mainframe. In 2000, I moved into a sales position for hardware sales in Europe, then corporate strategy in Armonk, and for the past three years I have been leading the mainframe business.

TPM: What is it like being the general manager of an IBM division? How is that different from being a systems engineer or doing sales? Is being a general manager a lot harder than you thought it was going to be?

EC: First of all, being general manager of the mainframe business is beautiful. The mainframe is the DNA of IBM. Being the general manager is like playing the church organ: you play with your hands, you play with your feet, you sing--it is a very powerful instrument if you know how to play it.

But all joking aside, the mainframe is the heart of innovation at IBM. That is the most fascinating thing. And it usually ends up that with the mainframe, we are going into uncharted territory. When you manage the mainframe business, there is very little you can copy. Other companies can say they want reliability like the mainframe has, or to be able to handle the transaction loads of a mainframe, or to offer the same level of I/O sophistication as the mainframe. My conversations are different. I very often sit down with customers and they tell me what they need, and most of the time, I do not have an obvious answer on how to give them what they need. And I have to go back and ask the engineers to develop what these companies want.

One way to look at the position of the mainframe within IBM and why Systems Group has come together with all of the platforms is that this is a crucible of innovation that very few other markets can give you. Look at it this way: Intel has the fountain of PC innovation; we have the fountain of innovation for large systems.

TPM: The pSeries Unix server line from IBM has gotten a lot of press in recent years because of the performance increases that IBM started bringing to the market in 2001. By my analysis, the growth in the performance for single-system images for the System/390 and then the zSeries and now with the System z9, has been just as large, and you have really kept pace with the pSeries in terms of growth. Is this something you plan to keep on doing with the mainframe as you move into the future? Or is there a limit to the size of a single-system image that customers need on a mainframe? IBM's mainframes have Parallel Sysplex clustering that is arguably a lot more sophisticated than the clustering available on high-end Unix boxes today. How big do you need to get these things?

EC: I don't want to be too sophisticated about this, but first you have to differentiate a little and talk about what "big" means. We originally thought a 16-way zSeries would be plenty of power, but the reality is that we had to make available a 20-way, a 24-way, we do 32-way now. In the labs we have 64-ways running, and next month we will do 54-way, but that won't be a single-system image yet. Why do we extend the mainframe like this? Because we need to know that we are able to scale. And we are able to do this. We have a very solid system that does not show signs that it cannot scale.

You asked a slightly different question, though. Do you need this? And I think that you will be surprised over the next three to four years about how the topologies will change. When I say topologies, you just named one, and Parallel Sysplex is the leading scale out, large systems topology, and we continue to invest in that. On top of that, though, you will see us do things like the zAAPs [zSeries Application Assist Processor]. If you think about these Java virtualization engines, in reality they are a virtualization of a whole application tier, and so I need a single-system image that can mange many of these, and I would submit to you that Java acceleration is only one of the accelerations that we have in mind. There might be other stuff coming our way that would make us think of extending the topology for managing other workloads with specialized engines.

That is why we will continue to deliver mainframe performance, and if anything, you will see us accelerate the performance.

TPM: Are you going to get that performance through a mix of SMP-style scale out as well as adding more cores to the central processors.

EC: Yes, over time. Here's the way I look at it. Mainframe computing is a paradigm where all of the resources are virtualized, of mixed and logically integrated workloads, and of integrated management.

The people in Poughkeepsie have a saying, a shorter version of this philosophy. They say: Everything outside the box is a defect. So if you look at what we announced with the System z9, in which we said cryptography and key management had to be centralized and built into the systems, that is an example of this philosophy. Why? You could do cryptography with an external device. But if you then try to scale it, you would get an additional cryptographic tier between your CPU and your peripheral systems. Do you know what that drives in management? What do you do in 15 years when you need to find the key with which you encrypted something on a particular external device?

What did we do? We took cryptography and compression and we built it into the processor because otherwise you don't get the performance. On top of that, we created the secure key cryptography card that is still in the system. The way I think of it, the mainframe is a network of processors that behaves as if it were one single image, and the amount of processors that we are going to put into a mainframe will be as many as are needed and the technology allows us.

TPM: Is IBM going to be able to mix zSeries processors and Power processors to do this?

EC: Sure. Let me give you an example. The cryptographic card that resides in the I/O slot in a z9 has a Power processor on it. It is transparent. When you say mixing processors, you have to be specific about whether or not they are tightly coupled or not.

TPM: I meant relatively tightly coupled back in the processor complex.

EC: I like to talk about whether or not the memory is coherent or not. If you want memory coherence, then you need the processors to have the same instruction set. For specialized work, like I/O acceleration, TCP/IP acceleration, cryptography, and so forth--I don't know what is going to happen with XML, but XML is one of those workloads that will lend itself to acceleration--there are many architectures blossoming, and I don't have a technical answer on how we are going to do it, but I can tell you what it will look like when we do it: It will be integrated and it will be transparent to the customer.

I have an example of this. The cryptography card on the prior mainframes was based not on a Power processor, but a chip from another vendor. And we changed it under the hood without anybody noticing. We have some history with that, you know. You are familiar with the iSeries, right?

TPM: Somewhat deeply, yes. [Said very wryly.]

EC: That's what I read. [Laughing.] Just think about the architecture changes we did there.

One thing you can count on for sure. Since we have built the very characteristics of the mainframe into the hardware, we will always have our own CP [Central Processor] and instruction set.

EC: We are in the same timeframe. To me there are three big issues that we are going to tackle with future mainframe processors. First, the single-thread performance. Second, we are going to look at the potential of zSeries processors for multicore designs--extreme multicore, if you want. The third issue is single threads versus multithreading. We are looking at all of these technology options, and this is what takes out into 2008, 2009, and 2010. It is exactly the same discussion you have with the Power guys, since after all, it is the same chip design team.

When we talk about mainframes, we rarely talk about processor generations, although we clearly do have processor generations. When you talk about Power, it is always Power versus Itanium versus Opteron. With mainframes, it is different.

TPM: Well, it is only different because now IBM is the only supplier of mainframe processors. Back when Hitachi and Amdahl were out there making and selling mainframes, we talked about the processors. When Hitachi's Skyline came out, there was a lot of talk--just like there is in other markets today.

EC: Those were working within the same mainframe computing paradigm--everything virtualized, and it was all running the same software. And when we talk about SMPs in the Unix world, even if they are hardware SMPs for 128-way processing, in reality the deployment pattern is for customers to put multiple, small Unix images that are shared on the same hardware. It is not the same development model as a Parallel Sysplex with a many nodes, which is actually a single system, and you manage it as a single system.

TPM: I would agree. Unix has a ways to go, and while Oracle 10g might look and smell a little bit like Parallel Sysplex, it is only doing it at the database level.

EC: The integration features you have within the mainframe--the two-phase commit with multi-rollback, the level of automation, and integration that you have inside--it is a different DNA entirely. And that is why customers stick with the mainframe.

TPM: A lot of that customer support has to do with the facts that they have lots of CICS code, DB2 experience, lots of COBOL applications. These applications are not easily portable, and equally important, even if you could move them, there is usually no technical reason to move them: They work and people understand them.

EC: I am agreeing with you. There is a wealth of sunk investment out there in mainframe applications that work perfectly and are perfectly stable. But if you look at our workloads and what's really driving the mainframe in the past few years is the fact that we opened the system up.

Consider this. At the end of 1999, we had a mainframe system that was 31-bit, CICS, COBOL, and did SNA. It was an isolated system. In 2000, we moved the hardware to 64-bit with the zSeries 900, then we added TCP/IP and then we figured out that Java and Linux have to come on the mainframe. And this is the path that we have been on ever since. If you look at DB2 Version 8 and at WebSphere V5.1 and higher, you can take code developed for these platforms on any other system and deploy it on the mainframe.

Here's how I look at it: I want to be the deployment platform of choice. I want customers to deploy their stuff on our boxes so they get the quality of service, systems management, and security that they deserve.

And had we not opened up the mainframe, we would have been relegated to a niche. If you look through Gartner and IDC data for the past few years, we have taken 17 points of market share. We have been growing our revenue market share. The mainframe has shipped more capacity in the last four years than IBM shipped in the prior 36 years, and that only happened because we opened up the system. This would not have been possible with COBOL alone.

TPM: These things have caused a resurgence in the mainframe, no question about that. Sales picked up, and recently leveled off, and then declined a bit as the System z9 transition began. Will Linux, Java, and maybe in the future XML--and all these other workloads that might get added to the mainframe--be enough to keep the revenue stream for mainframe hardware growing? Is there enough demand? Are we at the front-end of that wave?

EC: I think there are two factors that are right now helping us. The factors that you named--Java, Linux, XML--will help. Any standard you name, we are going to have support for it. We want the mainframe to be the hub for services-oriented architectures, and we are ready for that. The other factor is this: I think the increasing complexity that companies are facing will help us. Large enterprises have to deal with their disconnected and distributed systems. If you go into a very large establishment, very often they have 30,000 servers. They really do. And if you think of just keeping them going, that's one thing. But it becomes a nightmare when you think about security, resiliency, archiving--frankly, just knowing what all of these systems are doing is a nightmare. I am very optimistic, especially since other vendors are helping us because their solutions are so complex.

Look, I just installed my first network-attached storage array over wireless at home, because my girls want to have access to photographs and their iPod music, and they are clogging up all of the disks in the PCs. In my home, we can still manage without the mainframe.

TPM: But you could make yourself a small one, if you really wanted one. You have that power. The rest of us don't.

EC: Well, you have to remember: The IBM finance guys are watching over my shoulders. But it sure would heat my house, wouldn't it?