Dr. Jai Menon, CTO and vice president, technical strategy for IBM's Systems and Technology Group, visited us recently and shared his vision on what banks' data centers will look like in the future. Guess what? It's not all about x86 servers.
BS&T: I read that you helped invent RAID and hold 27 patents on RAID. Have you always been into storage?
Menon: Initially I was a database person in university. But when I joined IBM I was attracted to storage. I broadened out in the last 10 years or so. In 2001 I moved out of IBM Research. I wanted to move into the product divisions and pull new technologies faster into products.
BS&T: Which types of products?
Menon: I'm in the Systems Technology Group which handles all of IBM's servers, including the X series, P series, and mainframe or Z servers, as well as storage devices and systems. STG also has a microelectronics division, so we make microprocessors that go into our servers and ASICs and chips that go into things like smart phones, network switches and game consoles. I tell my son his game console has an IBM chip in it and he doesn't believe me. I've told the marketing guys they should make an "IBM Inside" sticker. Operating systems are also part of our division, including Linux, AIX (the operating system that runs on Power), and zOS (the operating system that runs on our mainframes). Linux also runs on our mainframes.
BS&T: So you're in charge of a lot of the high-test data center equipment that banks use. A lot of banks still have mainframes. I virtually attended the System Z announcement in June.
Menon: That Z series launch was an important start in some new directions.
BS&T: Are we seeing a real renaissance of the mainframe?
Menon: Here's the way I think of it: one key thing that got announced along with the Z196 was the zEnterprise BladeCenter Extension (zBX) companion technology that can hold blades of other architecture types like Intel x86 and Power blades. The reason that's important is because big financial institutions have all the architectures -- they have Unix, Linux, AIX, Windows, ZOS. They're big systems, like a quarter million processors, 1,000 switches in a data center, five million square feet of space. The direction we want to go is what we call Flex IT. How do I take a customer like that, that's very diverse, has lots of different architectures, and create a system that integrates all their hardware, software and application needs without their having to migrate their existing application or buy many different kinds of equipment. The vision we have is technology that will allow us to integrate all of that in a much smaller space.
How does a bank customer with COBOL programs take advantage of new offerings that are x86 based? You can move just the subset of your applications that are already on x86, those will be easy to move, but what about the other 70-80% that are not x86 based? Migrating the applications is a huge job that nobody wants to do. Our vision says we'll give you that same integration, even better density and a smaller data center footprint with incredible power improvements, but you don't have to migrate your applications because in this architecture, your Z apps can run, your x86 apps can run, your Power apps can run, we can carve out so much of X, so much of P, so much of Z. In fact, when you buy the machine you don't even think of it as a Z machine or an X or P machine. You buy the system, then you decide later how much of that will be X, P, or Z. You add more x86, more P, more Z, more network, more storage and memory capacity and flexibly allocate it amongst your apps.
The System Z launch was the first step with that vision. Because with the Z enterprise and ZBX, you now have this multi-architecture kind of thing with Z, P and x86 blades. It's an important vision because it's driven by what customers are asking us. It's not saying I'll give you this fantastic new thing, but you have to spend two years to migrate all your apps to it. That's not being customer driven.
It's not just about the past or legacy you're bringing along, although that's hugely important. There are billions of apps out there worth trillions of dollars, they're written in COBOL, Assembler, there's so much value in those existing apps and they'll be with us for many generations. You have to take that with you. But you don't yet know what the future will bring, there's still so much yet to go in terms of technology and architectures and what's happening in chips. The future will have new architectures. You've got to have architecture or system that supports not just legacy but also what you don't even know is coming. You can't just lock everything in and say the answer is one architecture called x86, which was invented in the PC era of 1980, that's it, we're never going to innovate again. That's foolhardy.
BS&T: A lot of the things you've been talking about are also goals of large bank technology leaders and the Open Data Center Alliance. They talk about giving vendors roadmaps that would let large user companies plug and play different technology components in their data centers and mix them with external cloud services. Would you work with a group like that?
BS&T: Are you working with them now?
Menon: I'm not sure. We have 220,000 technical people at IBM. I know we're working with them but I'm personally not plugged in to that. But that's the picture I'm painting, too. This vision says you've got to be able to support multiple architectures. It's kind of like the way SOA is there for the software world, in some ways this is SOA for the systems world. You can plug and play different kinds of systems and architectures.
BS&T: Are standards and cooperation with Intel and others required?
Menon: Z already supports Intel and Power, we're already demonstrating the first steps here. We can innovate within that in terms of interconnects that allow you to connect these things with high bus speeds, so we'll compete on all those other dimensions, we'll compete on how seamless we're going to make the integration of these things. When you run a multi-tier workload, you come in and tell me, this is my workload, this is what I need, and I'm going to have the system decide which pieces of that will run on the Z, the P or the x86, how much memory and how much storage is needed, so the customer doesn't have to decide, I'll write this for Z and do this for x86 and this for P. I'm going to try to make that as seamless as possible.
BS&T: Do you have that kind of workload management capability now?
Menon: We do for the Z enterprise. The first version of that is called the Unified Resource Manager. The longer term vision would be that you don't have to have a mainframe to buy one of these integrated systems.
BS&T: I would say that's what a lot of large banks are trying to do with their private clouds. In the past they've used software like Platform Computing and Data Synapse to help them allocate work among servers, running their high-frequency trading applications on their dedicated high-performance hardware while letting less time sensitive applications run on virtualized x86s. They would have to make those decisions. Banks continue to build on that, but I think they tend to call it internal or private cloud computing. Do you call this cloud computing?
Menon: I think of this as an architecture that would be the infrastructure you need to create clouds. A lot of people think that clouds only need one type of architecture and that's it. That's because nobody has yet built the kind of machine I'm talking about that will support different kinds of architectures within a single system. This kind of architecture would be perfect for clouds. Most of these things start out as being private clouds, as people virtualize, then they say what I'd like to do is put a self-service portal around it as people use this and allocate things for themselves, that evolves into a private cloud. A lot of why clouds migrate toward using single architectures is we haven't yet given them the kind of architecture I'm talking about that externally simplifies things so that you specify the workload, the system carves out what you need.
BS&T: As you pointed out before, different applications are written in so many different languages and vendors have proprietary APIs, we're not at a point where you can just pluck an app off your mainframe and throw it on an x86. What is IBM's vision for banks that have these old applications that still work and would be hard to rewrite in a amore modern language. Is it basically go from your old mainframe to a new mainframe?
Menon: It's putting a wrapper around these old applications to allow them to be accessed by web services. It's a SOA concept, you put a wrapper around it, you make it accessible by a web server, that's the way you modernize these things.
BS&T: What are you doing that's new in storage?
Menon: When you think about the different layers of storage, you have solid state disk, spinning disk, if the middleware knows a lot about which kinds of data might be attractive where, we can use software and algorithms to do that placement. We have a new offering Easy Tier that helps you figure out with software algorithms how you're using your data and puts the more actively used data automatically on solid state disk. It's seamless to the customer unlike other offerings where as an administrator you have to decide. Customers have smart admins, smart humans who know which data sets are used a lot. But we have MIPS to help you figure that out, you don't have to use human smarts to do that.
BS&T: You used to have hierarchical storage management years ago, now it's evolved --
Menon: Into something more flexible and self-managing. So you don't have to have user intervention.
BS&T: In trading you still have data that must move in microseconds, then you have archived data that doesn't need that kind of response. This kind of speed is becoming more important to traditional banks as they provide real-time transaction synchronization across mobile banking, branching banking, ATMs, etc.
Menon: IBM has introduced new technology called Infosphere Streams. It's a new, parallel systems architecture that lets you do a lot of analytics on the fly as the data is coming in. It's important for things like the trading floor.
BS&T: This type of technology is often called complex event processing in the financial services industry and it is used a lot in algorithmic trading. Do you have any banks using Infosphere Streams yet?
Menon: Yes. We did a proof of concept with TD Bank where we achieved a 27X improvement in options trading. You'll see greater use of field programmable gate arrays and accelerators for some of the processing you're talking about.
BS&T: A lot of banks use those but would prefer to be able to use standard hardware, they don't want to have to do special programming for specialized chips.
Menon: We can do things to improve the performance of an x86 processor. Our X5 architecture lets you have two to four times as much memory on an x86 server. It's additional silicon that attaches to the Intel processor using a standard bus. Why is that important? If you go to customers and ask them what is the average utilization of your x86 server, they'll say 20-30%. You'll say, what stops you from making 60-70%, why can't you consolidate even more workload? The answer is, there isn't enough memory on the server, because I've got to give 8 gigabytes to each virtual machine, so to run 30 virtual machines I need 240 gigabytes.
BS&T: Is this chip based memory?
Menon: It's standard memory DIMMs. We can put 128 standard memory DIMMs on a server. Then you can run Linux, Windows, vmware, kvm, or Xen; you work your way up and at each level it lets you have what you want.