Modularized system componentry - a tour of tomorrow - part 1 of 4

Today, I'm starting a multi-part blog on system architecture modularity in the world of tomorrow - this is part 1 of 4...  In case you haven't noticed, the technology world seems to be slowly but inevitably circling around a revolution in how we build and manage hardware.  And it signals a significant shift away from our distributed systems architecture driven from the Personal Computer revolution of the 80's. The next revolution is going to become one of increasing hardware modularization, commoditization, and virtualization.  We're on the verge of new technology interconnects and systems that will revolutionize the way you view hardware.  Take a look a few years down the road with me, at a few of the different areas where convergence and modularization is happening, and what it is likely to mean to the industry:

Processor Modularity

I'm predicting someday soon, we'll be working with a set of modular processor sub-components, configurable in a plug and play fashion.  We're going to see this happen through distribution of more and more features across the motherboard, and into and out of processors as the leading processor vendors fight with I/O and functionality struggles in differentiating their products.  The first indicators of this happening are around GPU (graphics processing unit) functionality being used to move floating point requirements off of the CPU for specialized applications.  This breaks the dependency upon over-provisioned CPU's for these applications just because they require floating point, and lightens the burden of overly expensive and unused processor components (floating point units) on the rest of us.  Take a look at what PeakStream, AMD and their ATI acquisition and Torrenza platform, cell processing, and other initiatives are doing in this area for High Performance Computing.  A good overview here from Ars Technica.

We're coming up to a point where it looks like processing may be highly distributable, linkable, and reconfigurable within a system, and this redefines the scalability of a system.  When suddenly you can grow the capabilities of your system by expanding individually the shared memory, floating point capabilities, specialized processing capabilities (think about I/O specific processing such as TCP Offload Engines), or core processing capabilities of a system over common high speed buses such as hypertransport 3, PCIe, infiniband, or the next generation interconnects, then the scalability of systems such as Fabric7 look small, but Fabric7 is nonetheless a precursor to this type of increased flexibility.

So this type of modularity and scalability in local systems will challenge our current server based computing paradigm tomorrow.  Today's consolidation approaches will seem like poor work arounds compared to what might be here tomorrow in the way of system flexibility.  But there's even more happening in this area today, which will further extend this modular system approach in the future.  And this in the next post...