As you might have noticed, this is not the typical "Letter from Bill." It's Bill Vass here, and while many of you might be familiar with me as the CIO of Sun Microsystems, I've also recently become the executive sponsor of the Sun Inner Circle newsletter.

Before we get to the substance of my inaugural letter, I'd like to thank my predecessor, Bill Howard, for his thoughtful commentary in this venue over the years. Bill has used this platform to offer CIO-level insight into the latest industry trends and Sun announcements — a tradition I hope to continue. Suffice it to say, I have some pretty big shoes to fill.

By way of introduction, I invite you to check out a representative sample of my published thoughts. I regularly post to an external blog where I share my day-to-day musings about the latest developments at Sun. As far as the industry is concerned, I've written about my Top Ten Rules for IT, which is a list of IT essentials that explain how to succeed with new technology. You can also listen to a podcast in which I discuss my strategies for consolidating and simplifying IT architectures and talk about some of the cool things we are doing here at Sun.

A Niagara Falls of Throughput

This edition of Inner Circle coincides with the release of the breakthrough chip multi-threading UltraSPARC T1 processor, internally code-named "Niagara". The feature story in this Inner Circle does an excellent job of explaining how the UltraSPARC T1-based servers are poised to streamline Web services. In this letter, I'm going to focus on the implications of the UltraSPARC T1 processor from a CIO-level perspective.

First, it might be instructive to discuss why we decided to code-name the new UltraSPARC T1 processor "Niagara." The "Niagara" processor shares its name with a famous North American waterfall in western New York and Ontario, Canada. While it might seem unusual to name a processor after a waterfall, the name "Niagara" offers an analogy for the promise of the new chip.

Niagara Falls — like Inga Falls in the African Congo and Khone Falls in Laos — is far from the world's tallest waterfall. Angel Falls in Venezuela, Yosemite Falls in California, Tugela Falls in South Africa, and Browne Falls in New Zealand reach much higher into the sky. But, while the towering falls move a comparatively thin stream of water very quickly, Niagara Falls, Inga Falls, and Khone Falls are much wider, and consequently, move a much larger volume of water in a comparable time frame. So it is too with the Niagara processor — there are chips that run faster than Niagara, but no other processor on the market offers as much throughput.

How Does the UltraSPARC T1 Processor Dramatically Improve Throughput?

The answer lies in the chip design. With today's Intel-based chip architecture, each time the CPU speed doubles, there is only an incremental increase in performance. And, in some cases a slower-clocked Pentium M processor outperforms a high-frequency processor¹ :

As this example highlights, despite the 1.4 GHz clock speed disadvantage, the Pentium M actually outperforms the Pentium 4. Why? Memory is a bottleneck. A high-performance Xeon server configuration today houses a 3600 MHz processor connected to DDR2 memory operating at merely 400 MHz. Every time the chip accesses main memory, the processor must wait an average of 100 nanoseconds for the data to be brought from memory to the processor. For a complex superscalar processor like the Xeon, the time it waits for memory creates a very power inefficient and low throughput processor.

As a result of the memory bottleneck, designers typically devote a large part of the chip to complicated architectural techniques with minimal and diminishing returns. In traditional architectures —- where only a few percentage points of performance improvement is possible — these techniques provide modest returns. With a radical throughput architecture, the complicated techniques pale in comparison to the performance gains now offered by the UltraSPARC T1 processor.

The UltraSPARC T1 processor is an eight-core chip. Each of the eight processors is able to handle four distinct threads. So, as far as software is concerned, the UltraSPARC T1 processor is a 32-way chip. In other words, an UltraSPARC T1-based server acts like 32 independent processors working in conjunction, instead of a single processor running really fast.

What Are the Implications of UltraSPARC T1 Processor for CIOs?

Universities, government, and Sun (and more recently IBM and others) have a long history of sharing intellectual property and contributing to and benefiting from "open source" and open standards. The idea of sharing with communities has gained tremendous momentum. Free technology is something expected more and more with the understanding you will pay for service and/or support on a subscription or per use basis data centers, where volume is of crucial importance for the effective delivery of Web and database services.

It is no secret that to realize perceived cost savings, enterprises have deployed large volumes of inexpensive chips at the edge of IT infrastructures. For example, it is common to see scores of commodity Web servers supporting large Web applications. The trouble with this strategy is that it is still expensive to buy a lot of servers. It is difficult to manage the servers. The servers take up a lot of space. And it costs a fortune in power to cool the servers. As a result, today's CIOs are running out of data center space, while paying ever-increasing costs to procure, manage, and run Web application infrastructures.

For multi-threaded applications — such as Web serving, application serving, and lightweight database serving — where CIOs have traditionally racked up large numbers of inexpensive servers, the UltraSPARC T1 processor offers the opportunity to replace many of those servers with a single server that uses much less power. Of course, the UltraSPARC T1-based servers are not for every application. There are still areas where clock speed matters, such as in high performance computing. For applications that require fast floating-point calculations and for single-threaded applications, a Sun x64 or a traditional SPARC might more sense. But for multi-threaded applications, the UltraSPARC T1 processor is specifically designed to address the needs of today's data centers.

The UltraSPARC T1 Processor Curtails Costs

By offering more throughput in a smaller package, UltraSPARC T1-based servers herald a number of cost savings. In addition to the direct cost savings brought about by server consolidation, UltraSPARC T1-based servers can help reduce indirect costs. The most obvious example results from having to maintain and manage a significantly smaller and more standardized data center. Also, by consolidating server infrastructure, UltraSPARC T1-based servers allow enterprises to reduce the number of software licenses needed to support an application.

There is another important area where UltraSPARC T1-based servers are poised to save enterprises money. With UltraSPARC T1-based servers, enterprises can dramatically reduce licensing and support costs. Compared to Windows, the licensing costs alone offer a significant savings. And while the Solaris OS and other open source operating systems like Linux come with a free right-to-use license, most CIOs opt to pay for support contracts for anything they put into production. With UltraSPARC T1-based servers, enterprises pay support costs for significantly fewer CPUs than with an Intel-based architecture, which can cut costs several times over.

The UltraSPARC T1 Processor Promotes Energy Efficiency

With UltraSPARC T1 processors, enterprises need fewer servers to support a typical Web application, and that means fewer servers to cool. And by reducing the amount of server space needed to support an application, the UltraSPARC T1 processor can help shrink data center real estate, which means less space to cool. While those are the obvious energy efficiencies created by the UltraSPARC T1 processor, the chip itself helps deliver cooling cost savings by forgoing the futile emphasis on clock speed prevalent in today's Intel-based servers.

For example, the Sun Fire T1000 server typically uses nearly 18 times less power than a comparable grouping of previous generation Xeon-based servers, based on customer testing. For the workload supported by 500 Sun Fire T1000 servers over three years, this adds up to a colossal $7.3 million in power and cooling savings, ignoring the further savings from reduced space².

The UltraSPARC T1 processor can dramatically lower cooling costs, but you don't have to take my word for it. As Rene Wienholtz, CTO of STRATO AG, the third largest Web hosting provider worldwide, relates, "After doing a TCO calculation, we see fantastic opportunity with the T2000 to reduce our power and cooling costs by replacing some of our current systems. This CPU simply doesn't generate much heat."

Designed for Today's Enterprise Data Centers

One question I hear a lot is: Enterprises spent a lot of time and money switching from SPARC to x86, so why would they move back? To which I answer: Why wouldn't they? First, for people who moved to Linux, it is easy to move back to Solaris. Second, UltraSPARC T1-based servers will save enterprises a significant amount in hardware procurement and maintenance. And, UltraSPARC T1-based servers can dramatically improve server utilization and reduce cooling costs.Also, there's no need to re-write applications when you switch to UltraSPARC T1-based servers. Making the switch simply allows you to run your applications more efficiently.

Among the customers who tested the UltraSPARC T1-based Sun Fire T2000 server is the College of Oceanic & Atmospheric Sciences, a research college at Oregon State University. When asked to assess the performance of the SunFire T2000 server, Chuck Sears, the manager of the Research Computing department, says, "We were extremely pleased with this system right out of the box. It is an impressive unit in how it processes and serves up business functionality, such as Web services and messaging infrastructure loads.

"Early indications with a very small amount of performance enhancements to our code bases are yielding 3 to 4 times increases over some other of our systems," Sears continued. "Our internal code bases are running so fast that we were not sure if they were actually working correctly. It appears on our initial test cases that this machine will FLY."

Just like the waterfall that bears its name, "Niagara" servers offer comparatively massive throughput. In other words, while Yosemite Falls might look pretty and seem fast, Niagara Falls does all the work.

Regards,

Bill Vass
Chief Information Officer
Sun Microsystems, Inc.
cio@sun.com

¹ SPEC and SPECint are registered trademarks of the Standard Performance Evaluation Corporation. Results taken from http://www.spec.org on 11 November 2005. Dell Precision Mobile Workstation M60 (Pentium M 755) — SPECint2000 1541; Dell Precision Workstation 360 (3.4E GHz Pentium 4) — SPECint2000 1484.
² Sun customer testing showed 6x speed-up for a single UltraSPARC T1 server when compared to their existing previous generation dual processor Xeon systems. Each Xeon system consumed 585 watts versus the Sun Fire T1000 server.