point
Menu
Magazines
Browse by year:
Caution, Intel Architecture Ahead
Monday, April 1, 2002
Where is Itanium headed?
We’ve taken a very focused approach. Driving an architectural change in the ecosystem is a tremendous amount of work. With IA-32 we have a base that’s in development for a decade plus. What we are doing with Itanium is based on where the technology is hardest and where the maximum benefits are we are focusing on the leaders in those industries and focusing our energies there.

A good example would be database. Clearly Itanium, and its architecture and platform is very conducive to supporting databases and delivering transaction performance. So Microsoft, Oracle and DB2 make up probably 70 percent of the database market. So we’re working very closely with them, relative to their databases.

We’re taking that same approach for some of the other targeted application areas. Overall, it’s really going after a market that we haven’t participated in very heavily. It’s going after a huge business that we can’t go after with our current IA-32 platform.

So how do you see Intel fitting into this market?
If you look at the overall revenues from a server standpoint, Intel architecture-base platforms enjoy maybe 35 to 40 percent of the market. 60 percent still resides on a RISC processor whether that’s IBM, Sun or HP. Certainly with our engagement with HP and our engagement with Compaq we know we will start having channels and access into that 60 percent that we are just not participating in.

So in a way we will be a disruptive technology in these RISC areas. We are going to be comoditizing to some extent the markets that Sun has been going after.

What are the key technology developments that are emerging in enterprise computing?
One trend would be blade-based computing. We went from a pedestal product to a high-density rack product, driven by a space and power constrain. The more power you crammed into a cubic centimeter the more revenue you got out of that space.

With blades you’ve got another set of vectors being addressed. If you look at the bladed architecture today (and there are a whole bunch of different companies that are investing in this particular area) it’s the disaggregation of computing, storage and networking. If you look at a traditional server, it will have a single motherboard, storage, some level of network and gigabit connectivity outside the box. That’s your server and you connect 30 of those or however many you can fit into a rack and you have to manage all of those separately. You get fantastic performance per cubic centimeter, but the manageability costs can be overwhelming depending on how large your database is. So you have companies developing bladed architectures. You have CPU blades or compute blades that are separate from your storage and your network and I/O. Think of it almost as a virtual pool of compute power, networking power and storage power. Software is the key to developing these bladed architectures so with software you can create virtual servers and utilize whatever resources you want. You are dynamically managing the applications that are running on these nodes and dynamically adding more compute power. Say your number of e-mail users go up by 2x you can dynamically add more compute nodes to the virtual server that you have created for that surge in demand.

That’s a major trend. One aspect of it is total cost of ownership (TCO), another aspect is scalability. Bladed architectures are supporting overall better management, and an easier way to scale out.

Another motivation is that if you look at the RISC competition for Intel architecture, they are still advocating large 8 16 24 or 32 way platforms and they are providing partitioning capabilities. So Sun will come into an account and say you don’t want to mess around with 100 Intel servers, buy one of our 24 or 32 way platforms and I’ll give you partitioning capability to partition it into 100 servers and you can manage it more easily. If you look at that technology it doesn’t deliver all of what it says it will, but the bladed architecture is a response to some of that threat as well to try to ease some of the TCO.

How soon will the bladed architecture become prevalent?
You have people announcing products today. Compaq announced their “QuickBlade” platform earlier this year. Most of the OEMs are developing platforms. It’s here and now. You will see a lot of investments in this particular area around Intel architecture products in the next few years. So that’s a major trend.

Synergistic with that trend is high performance computing. We’ve got a lot of good examples especially around Itanium. The biggest one was introduced last year by the National Science Foundation — roughly 3000 Itanium processors being networked together through high speed interconnects — it was almost like a computing utility. That particular one is being used for scientific calculations. But I think that you’ll see more of those being deployed for business computing.

What’s your direct response to the criticism that Itanium has yet to take off?
It all depends on what ruler people had out. Relative to our expectations it is doing what we expected. Relative to past RISC architectural transitions it is doing far better. If you were to map it against an IA-32 processor as to how fast that would ramp it’s the wrong comparison because that’s got an ecosystem that’s already there. The best yardsticks are other architectural transitions of this magnitude.

What opportunities for startups do you see surrounding these new developments?
There are opportunities in blades and there are a whole host of startups in that category — not only on the hardware side but on the software side because the software is a lot of the challenge.

I think there are opportunities coming on the telecom side. The telecom industry is going through a pretty significant transition. It has faced about a $50 billion drop in annual spending. And that drop has forced everyone to get creative — to look at open standards and open building blocks.

So given this downturn in telecom, is Intel using its resources and balance sheet to jumpstart its efforts in the telecom market?
We introduced two carrier-grade products last year, based on Intel architecture. These are 1U and 2U density products, but they are geared for the telecom market, the central office, and some emerging applications like soft switches, VPN, voice-over-IP gateways. These things are designed in accordance with all of the telecom national equipment building standards.

There were multiple objectives with these products. One is that we kept seeing demand from telecom equipment OEMs to use standard products. They see lots of reasons. They saw the benefits of that from a technology treadmill standpoint and they couldn’t continue affording the proprietary presence. One of the motivations was that we saw Sun, with their carrier class networking product line was enjoying a lot of market success without any competition from the Intel architecture side.

Our communications group, in addition to all of the silicon that they are investing in, whether it is Ethernet, or the optical product, or even the network processor — again another standardization on the communications side — they are also investing on the system level. There is a standard called Compaq PCI — which is another blade-based standard. So we are developing blade building blocks that we’ll also make available to communication vendors.
si

Twitter
Share on LinkedIn
facebook