Can we hack open source #cloud platforms to help reduce emissions?

RedMonk runs two conferences a year – Monki Gras in London in late January/early February and Monktoberfest in Portland, Maine in early October.

At this year’s Monktoberfest I gave a talk titled “Can we hack open source #cloud platforms to help reduce emissions?” to an audience which consisted almost entirely of developers and the feedback I received was consistently positive. The slides for the talk are above and I’ll post the video as soon as it is available, along with a transcription.

My presentation was fairly straightforward – it was a call to action. I first outlined the problem – most cloud computing companies are not publishing data around their energy or (more importantly) their emissions. Those that are, are not publishing it in enough detail, or are publishing the wrong data (carbon saved, as opposed to carbon emitted). I also pointed out that energy use is not a proxy for emissions – the amount of emissions the important metric to track.

Why don’t cloud computing companies disclose their emissions? It is probably a combination of not wanting to give away competitive intelligence, not wanting to be viewed as a polluter, and there being no agreed reporting standards around this area.

Then I pointed out a quick recap of the year 2012 to-date with all of the wildfires, floods, droughts, temperature records, and unprecedented ice loss in the arctic. I know these aren’t solely caused by cloud computing, but it is a significant contributor (estimates from Gartner in 2007 put the amount at 2% of global emissions – and that number is highly likely to have increased since then).

The solution – I proposed a fairly straightforward solution. Hack the currently available open source cloud platforms (Eucalyptus, CloudStack and OpenStack), write emissions measurement and reporting patches. Get the patches accepted back into the core so that when the next update of the software is pushed out, the companies using the three platforms will at a stroke, have energy and reporting capabilities. At that point customer demand should ensure that they make this info public (or at least available to their customers).

I concluded by noting that adding emissions metrics and reporting to cloud computing platforms will reduce emissions – and then asking the audience “Ok, so who’s up for it?”

During the Q&A, Andy Piper rightly pointed out that it would have been appropriate for me to mention the Cleanweb Manifesto in my talk and he was absolutely correct. Next time I give the talk I will point it out and urge people to sign it.


  1. David Aikema says

    I’m not quite sure how effective your idea of pushing energy metering info the open source cloud platforms is likely to be. I’m guessing that even if the capability is there in OpenStack major public cloud providers (at least the less efficient ones) probably will have it disabled on their servers until some sort of critical mass is achieved. I’m not sure that that critical mass is there yet. Another issue with VM power metering that seems to persist in cloud environments would be the issue of who’s assigned responsibility for all the idle power – both on the server you’re currently running on as well as other servers in the data centre (given that you probably don’t have control over which machine your VM is run on). More or less you’d seem to be able to get the incremental power consumption of a specific VM’s actions, which you note doesn’t actually account for all that big a portion of total energy consumption. Even incremental power consumption, to a certain extent, will be influenced by the other VMs it’s sharing a physical machine with, perhaps posing privacy concerns and providing further incentive for those operating public clouds to keep the reporting functionality disabled even if it’s there.

    I’ve been involved in the development of an ISO-14064-2-compliant standard for estimating emissions and emissions reductions for the ICT sector (addressing concerns like the emission intensity of the energy used by the data centre) as well as having been affiliated with a follow-the-renewables computing testbed.

    One thing that those experiences raised in my mind in response to this piece is that the sort of emission intensity data sources you’d likely be dealing with in many instances is something like an annual average whereas the power mix tends to fluctuate a bit over time. Wind seems in the cases that I’ve studied to have highest output during night hours, grids with lower average emission intensities may fire up peaking plants during peak hours, etc. (Granted support for variation in emission intensity in reported data could be improved, but there are also some latent emission effects with ramp-up/down that would further confuse things). It’s also a reason why purchasing renewable energy certificates (as got Rackspace London a zero-emission score on Mastadon C for example) can overstate the case somewhat by ignoring the integration costs for that renewable energy.

    And a pet peeve that’s developed over the past few years: I noticed that you commented on Amazon not providing evidence to verify their zero-carbon claims, but I wonder to what extent we should step back entirely from the term “zero carbon” in favour of low carbon or 100% renewable. I already mentioned Mastadon C’s rating for Rackspace London, but that’s not the only data centre listed as having 0.00 g CO2e/kWh emissions. Looked at from a total lifecycle analysis perspective wind, solar, hydro, and geothermal all have non-zero carbon footprints – they’re just quite a bit lower than for their fossil-fuel counterparts.