Follow-the-sun used to be about who was working when - most commonly for customer support. Beyond a certain number of calls and customers, particularly if the topic is not complex, it is cheaper to have people working in prime time in their time zone than to run two or three shifts in one timezone. So Cisco System's first call centers emerged in Sydney and Amsterdam, so technical support would be at UTC -8 (Menlo Park,CA) UTC +10, and UTC +1 (modulo daylight savings).

There's a new use for the same phrase - computing tasks are moved to a different geographical location to take advantage of differential electrical power costs. Solar power and wind power are variable by nature. Being able to use them close to the generating source reduces the transmission losses from moving the power to where it is being used.

In iSGTW this week "Canada’s Green Star Network aims to demonstrate that by allowing the computations to follow the renewable energy across a large, fast network, the footprint of high-throughput computing can be drastically reduced."

Google already has the capability to do this, as a side effect of the overall engineering design for their infrastructure (characterized as "design for failures"). Newer data centers in some parts of the world can be built without chillers, so that cooling uses only the outside air. On the occasional hot day, the effective compute load for that location goes down, and bulk machine to machine traffic can be moved. This could be described as follow-the-cloud,to find cooling capability, rather than follow-the-sun to find power.

Moving computation to where there is cheap sun or wind power sounds like a 'why didn't I think of that' idea - until some of the necessary details are considered.

What's the actual thing being moved ? An application and all its associated data ? A virtual machine ? Using which operating system and hypervisor ? Greg Pfister has 4 posts describing hardware virtualization in detail - the memory management issues when moving a VM over a significant distance (and therefor over many milliseconds of time) bear consideration.

Bandwidth - Google owns a lot of its own fiber, so can run many 10Gbps (now, 40Gbps and 100 Gbps soon) streams on its WDMs. Greenstar will have access to the Canarie research network, which has 10Gbps on each wavelength.

Availability - Design for high availability system wide requires keeping spare capacity - for computing, for power, for connectivity. Designing a central control point rather than a distributed system seems easier, but doesn't scale well.

Measurement - technologies to measure where power is actually available, and models to price optimize its use, are in their infancy. Certain consumer and industrial use patterns vary consistently by time of day and time of year, but reacting on, say, 15 minute time scales at anything other than very coarse granularity (like turning up and down a gas fired power station) isn't something regulated utilities find easy. For computing load, Amazon AWS publishes 'service health', but not capacity.

These are just a few of the issues which will need to be addressed. Standards are in progress for moving data between 'clouds', whether private or public clouds. It is possible to quickly turn up computing capability at different data centers; since data centers typically buy power on long term contracts there is no variable price signal to use to affect a location decision.

Summary : relocatable computing to follow the sun or follow the wind could happen, particularly if the price of power stays high enough to encourage the necessary development, but there are many interacting details, in systems which are currently opaque to their purchasers. Developing Platforms as a Service (PaaS) to support relocation for general purpose computing is a long term project.

References

iSGTW article on the GreenStar network
Cybera (Alberta infrastructure research institution) description of its participation in and use of the GreenStar networks
Google infrastructure talk at Nanog 49
Hardware virtualization, Greg Pfister
AWS status reporting

Cunning Systems evaluates product and service ideas in computing and communications. If you would like to discuss an idea, contact us at info@cunningsystems.com