As an IBM Corp. Distinguished Engineer of IBM's xSeries servers, Tom Bradicich lives off of hand-me-downs and he's proud of it. Unlike other server companies leveraging Intel processors, IBM gets to leverage I/O technology originally designed for higher end Unix and mainframe systems.
In an interview with InfoWorld Editor in Chief Michael Vizard and Test Center Director Steve Gillmor, Bradicich talks about how a new data center phenomenon called network convergence is reshaping the data center and ultimately creating competitive friction between IBM and companies such as Cisco.
What's happening in IT organizations these days? It seems that a lot of them are morphing into becoming enterprise service providers for their customers and partners.We're seeing that a lot. We're still seeing a traditional three-tiered approach, but we're seeing a newer phenomenon that I call the server network convergence, where it's becoming a bit indistinguishable where this network begins and where that front-end server ends. It's happening both ways. Traditional network-geared companies that would provide you with a switch or router are building in functions that would traditionally be on a server. Meanwhile, we're seeing server companies, like us, looking at building what you would traditionally purchase from a networking company. I'm building right now an Ethernet switch into a server -- the first time in history that that thing is built in. The blade server phenomenon is driving a lot of that. The whole competition of function between the server and the network is very interesting. There are going to be many times where the blade phenomenon results in [fewer] separate switch sales because they come baked into the blade. This is going to be a fact. And the end-user enjoys the benefit of not having to purchase two things from perhaps separate companies.
What impact will that have on systems and network management?We're driving a single view, so that a single person can understand if indeed a server's down and what to do with the network. All that should be automated.What are the cost benefits of blade servers?It's going to be a lower cost if you can amortize fans and power supplies across blades. Another [benefit] is density; obviously, you can get more on a rack. Then there is configuration; you could put a rack of three times as many servers together in a tenth of the time. It take less skill to put a blade rack together than somebody connecting literally 312 connectors in the back and running into errors, fatigue, and that sort of thing. You can also dip your toe in a new function or even a new technology. Alternative memories and alternative storage can be tested without committing [to them].
What software will be needed to make all of this work?That capability will be in IBM Director, which comes with every server purchase. So when you purchase a blade server from us, which will be [in] the fourth quarter of this year, IBM Director software will include an automated provisioning feature. It'll include some workload management. One neat thing we just did was called Concurrent Diagnostics. Today a system running Windows must be taken offline to run diagnostics. That's the nature of the beast. That's down time. We've just come with a technology where we can run diagnostic activity while the system's up and running. That's not new to the Unix systems, it's not new to mainframes, but it is new to Intel platforms. There's also Project Oceana, which is a single project associated with deployment and dynamic real-time movement. Project Eliza is an umbrella project of initiatives within IBM across our server lines to provide self-management, self-healing, and self-configuring.
What role will Virtual Machines play on Intel platforms?IBM invented VMs. We're good at that. VMs happen to be a natural next extension. We have a relationship through VM Ware that let's us get VMs on the Intel platform. We not only have certified that software product, which you know is available to anybody on our servers, but we also have a joint development with that company to make enhancements to VMs. We will have machines running, for example, both Linux and Windows.
What role will 64-bit systems play?We continue to be enthused about 64-bit technology. We have a very powerful project called Enterprise X-Architecture on I-64 McKinley processors. It being a powerful processor, it has a lot opportunity for virtualization. Our vision is we have blades in the top, as in tier one; blades in the middle, as a two-tier offering; and on the bottom, non-bladed standard McKinley for the back-end database. So you have a three-tiered rack concept.
What's next in Intel server architecture?It looks like everyone is behind enhancing PCI with PCIX 2.0, which is brand spanking new. And after that, it'll be 3G I/O, and then InfiniBand existing as a means for inter-processor communication for clustering. That's a lower cost alternative for high-performance storage than fiber. Conceptually, it's a point-to-point channel as opposed to shared bus. [When] a bus is shared, the more you have on it, the slower it gets. The good news is that for the first time in history we have an industry-standard way of doing clusters. There's been a lot of activity in storage this year. We've seen some demos by third parties who hooked up through InfiniBand connections [and] DB2 clustering and have seen some good performance numbers. I expect that by the end of the year you'll see storage solutions.
Is the Intel server market consolidating or fragmenting?The complexity of segmentation [is] growing. We have 11 servers now in our product line; it'll be 14 soon. My competitors have a similar amount of [servers in their] portfolios. So even though the expansion as a whole, in volume, of the Intel server market has subsided some, specialization continues to grow.
How is Linux doing in server environments?If you looked [at] the numbers, although Linux is growing quite briskly, its share is still relatively small [compared] to a Windows. Obviously, we're going to continue to enhance it.
What's your take on the proposed merger between Hewlett-Packard and Compaq?It's a good time to be us. I can say that.
So what's IBM's core strategy in the Intel server space?We continue to have a lot of support for what we call X-architecture strategy, where we steal from the high-end and put in the Intel-based system. That's good for us because that's a distinctive edge we can have against our competitors. We don't just trickle down; we shamelessly steal what those guys have invested.