Archive for November, 2004


November 25, 2004

A good woman died today. No one you’ve heard of, but much loved by family and friends. Still very much the New England girl to the end, with kind and gentle words and amazing grace through years of illness and pain.

Elizabeth Eleanor (Handfield) Guidette. Aged 60. Rest in peace, auntie.

Calise’s Law?

November 23, 2004

While at SC2004, the High Performance Computing, Networking and Storage Conference, in Pittsburgh earlier this month I heard Mike Calise, president of Clearspeed make the following observation:

The number of people who say Moore’s Law is dead doubles every eighteen months.

Cute…and if you think about Moore’s Law as a statement of processor frequency trends rather than as an area law as it was originally stated, then count me in. At Sun, we’ve been talking about this for quite some time and others in the industry, notably Intel, are starting to sing our tune as well. The future belongs to chip multithreading–using our silicon budgets to replicate large numbers of relatively simple cores on a single die–rather than to large, complex boomer microprocessors that continually push clock frequencies ever higher. Why?

Because there is a perfect storm of sorts occurring in the industry that makes this a useful direction to pursue.

First, network computing workloads are thread-rich. Meaning that as more users hit your e-commerce site, the more need you have for independent threads of execution to handle that load. In the web tier, the app tier, and in the database tier. A technology that could handle bigger workloads without blowing out the space budget in your datacenter would be a Good Thing.

Second, consider the fact that memory speeds have not increased at the same rate as processor frequencies. Which means now memory is many hundreds of clock cycles from the processor. Not good for many workloads. A technology that would accomplish useful work while waiting for data from memory would be a Good Thing.

On the technology side, if Moore’s area law does continue to hold we will be able to put one billion transistors on a die in the not very distance future. That’s a lot of transistors. More than we need for any single processor core. A technology that would let us use these “extra” transistors for something more useful than just more cache memory would be a Good Thing.

And on the reality side, where complexity and quality are in a constant struggle and where time to market is critical, a technology that valued modularity and a “less is more” approach to processor design would also be a Good Thing.

This FAQ covers some of this in more detail.