Photo via Tipiro

It is interesting to see the recent commentary regarding the environmental effects of Cloud Computing, and I’m happy to see the dialog coming while the industry is still in its infancy.  If only the same could have happened with the automotive industry - what a different world we’d live in today!

As Reuven Cohen (CTO of Enomaly, a Cloud provider) writes,

“The general consensus says that reducing the number of hardware components and replacing them with remote cloud computing systems reduces energy costs for running hardware and cooling as well as reduces your carbon foot print while higher DC consolidation / optimization will conserve energy.”

But as Mr. Cohen and James Urquhart (Product Marketing Manager, Cloud Computing and Virtualized Data Centers at Cisco) agree - there is no proof.  There is no standard measurement or parameter that everyone agrees indicates that the net result is better or worse for the environment.  And so for now we’re really discussing the hypothetical and using deduction to point us in the right direction.

One point we can all agree on - as a civilization, we’re doing more computational work now than ever before.  Of course, this will only continue to grow, and exponentially. 

Think about the rate of increase in the number of people performing some sort of computation (for example, the 300+ million members of Facebook all uploading photographs and playing Mafia Wars) and the rate of increase in the amount of data to be manipulated (consider a 5 megapixel camera built into everyone’s phone, or everyone watching Avatar in HD on Hulu).  All the while, in the cloud, processors will be running algorithms to deliver the experience while constantly making adjustments as they dynamically navigate the trade-off between data size, connection speed, and client performance (processor and screen resolution).

The question is, are we more environmentally friendly doing all of this in a shared Cloud or on our own datacenters?

Mr. Urquhart’s reasoning takes us in a positive direction,

“I believe one thing to be true: the increased efficiency of the hardware components in most cloud data centers and the increased utilization of these components mean that we are almost certainly doing more work per unit of energy consumed than before.”

But on the face of it, I can only agree with part of this.  Indeed, greater efficiency built into the hardware is a good thing for the environment - the lower the power consumed and the greater the amount of work done per clock tick of the processor is good.  However, greater utilization of that hardware due to virtualization could take us in the opposite direction with respect to the environment.

Firstly, (as both Mr. Urquhart and Cohen agree) the more compute cycles available, the more we’ll use.  It’s what we do.  Therefore, increased utilization due to virtualization will actually increase energy consumption and impact the environment negatively in absolute terms.

But more subtly, how the compute power is delivered is actually more important in comparing a datacenter to a virtualized cloud.  Virtualization improves utilization, but does it improve system energy efficiency over the same workload?

Whether the hardware is virtualized or not - the processor still needs to tick.  In fact, through virtualization, the processor needs to work harder with more fetches from cache, DRAM, or the hard drive to deliver the same calculated result to the end user.  And so, on the face of it, greater utilization due to virtualization actually impacts the environment negatively…for now.

We need to go back to the hardware to turn this around.  Virtualized servers drive the high-end of the processor lines of the physical servers that support them.  The increased demands of compute power per square millimeter of the processor and parallelization/efficiency within them will drive overall energy efficiency of the system in the right direction.

One additional point to consider is data distribution.  The power it takes to deliver the compute power from the cloud/data center to the remote user has an impact.  A centralized cloud vs multiple distributed datacenters (from different service vendors) and their proximity to the users needs to be added into the equation.  If we’re talking about consumer services sharing a cloud, the impact of centralization will be significant and detrimental.  However, corporations using their own local datacenters on-site are much more energy efficient in data transport energy costs.

Clearly, the calculation is not straight forward, and when all’s said and done, it may only be interesting from an academic perspective.

Fortunately for all of us, being green is considered by many to be a tier one value proposition during the purchase process of equipment and services.  This demand will get baked into the product definition process and we’ll all move in the right direction.  One point that is without argument however - it’s a heck of a lot greener to work remotely on a cloud or datacenter, than either shipping CDROMs or getting on a plane and working on-site.

Posted under industry

There is of course still a lot of bluster surrounding Cloud Computing, and it will take some time before an agreed to set of definitions comes to light.  I particularly like the path that Phil Wainewright is charting though.  He’s been commentating on the space since the end of the 90’s and has a “no nonsense” approach I appreciate.

In an effort to simplify the discussion, a recent post of his, “When is a cloud not a cloud?“, begins to define different types of “Clouds” from the perspective of the original cloud - the internet.  He lists three different types - Captive Cloud, Public Cloud (AWS, the Xuropa Platform primary cloud provider), and Virtual Private Cloud (a “walled off” cloud within a Public Cloud, like AWS). 

The post is worth a read, but he starts to run aground on two points IMO.  Firstly, to isolate connectivity as the key defining characteristic is to emphasize what should be a “cost of entry” point, and glosses over real value differentiators (eg. resource management via API).  Secondly, he really does himself a disservice when he includes “Captive Clouds” as a type of cloud (even though he was trying to be ironic).  We already have a name for that - “datacenter”.  A collection of networked servers running something like VMWare behind a firewall is not a “cloud” of any description.  It is a datacenter with better hardware utilization than the datacenters of the 90’s.  That’s it.  To put the word “cloud” in the same sentence as “datacenter” reduces clarity rather than enhances it.

Posted under Xuropa, industry