The Cloud Blog

Intel recently showed an experimental 48-core processor and called it the “Single-chip Cloud Computer.”  As some folks have already written, we’re essentially getting to the point where data centers can be built on a chip, and this will fundamentally change (and enable) how we construct data processing environments over time.

Not so many years ago, it would take an entire data center rack (not to mention commensurate power, cooling) to equate to the computing capability of this single chip.  But, why call it the “cloud computer”?  Well, to make use of this density of capability, you really must embrace the tenets of cloud computing: high levels of virtualization, parallel and distributed processing, and reliance on software (instead of hardware) to “tie together” the end solution.  For example, instead of following traditional tightly-coupled cluster design theory (i.e. hardware cache coherency), Intel instead adopted an embedded high-speed 256GB/s router network between the cores.  Why?  It reduces system complexity, power consumption, and provides a more generic solution applicable to a wide range of processing environments.  Think of it as 24 dual-core servers networked together.

Today, cloud solutions require fairly robust compute environments to meet customer requirements: encryption, compression, deduplication, content rendering, etc. each consume compute and bandwidth resource to deliver value to a customer.  A well-run cloud environment can provide cost advantage because high levels of virtualization and automation permit high utilization levels.  With a processor like Intel’s prototype, the overall efficiency and capability of that environment is substantially improved.  For example, Hitachi’s Content Platform employs a multi-node architecture that allows scale-out of cloud-based storage to a customer’s requirements.  As data needs grow, additional nodes can be added to keep pace.  By incorporating highly-dense and integrated technologies like those from Intel, the solution can be delivered with less power, cooling, floorspace and enhanced feature set.  In the true sense of cloud, those efficiencies get passed to the consumer as lower costs for similar (or better) service.  By leveraging capabilities such as the embedded router, data processing functions can be efficienctly brought closer to the stored data footprint (think content rendering, real-time video format conversion, virus scanning, capacity optimization…)

Now it’s left to the imagination what can be done when multiple 48-core CPUs are aggregated together…