One of the expected outcomes of industrialization is the ability to use standardization/consistency and scale to drive increased returns [definitition: “increasing returns to scale“]. Electric utilities are well known to be more resource efficient than home generators – higher yield & lower waste. These efficiencies typically come from both demand concentration/periodic demand complements as well as the supply side engineering for improvements in yield – this effect and affect are relevant to our Data Centers.
Demand Concentration and Scale Economics
Efficiency Innovations & Utilization Efficiency
Most businesses are periodic: A trading firm is most active during the daily trading window, retail is busy during holiday shopping and return to school, and online streaming services after work-hours. As a result, an optimal mix of workloads for a multi-tenant data-center could support both the 9a-6p work of a business and then an online streaming service from 6p->2a, and then maybe some fill in work between 2a-9a with pricing calculators, rendering projects, monte-carlo analytics and the like. Though not necessarily perfect compliments, modern cloud providers have a wide mix of consumer and business clients which gives them an edge vs. a [single] corporation’s data center
On a per workload basis, improved elasticity is being built into Enterprise workloads; a natural compliment to the increasing elasticity of current public cloud consumption and pricing. Thru more aggressive use of platform services (e.g. Kubernetes Clusters w/ Docker, Open Shift, Pivotal Cloud Foundry, etc..) enterprises are able to much more rapidly develop, deploy and scale [up and down] to meet the needs of the business, with some achieving a 50% reduction in bills [spend]. For novel or substantially re-engineered services, a true shift to Functions (serverless) can get to savings rates up to 90%. Both PaaS and FaaS present power usage savings inline with the cost savings rates: functions are the most green ($ and power efficient).
WINNER = Public clouds for increased returns to scale economics!
Now, let’s talk a little bit about yield. Yield is really a marker of raw material/resource conversion to useful product. Given the global conversations about greenhouse gas emissions associated with power generation and the increasing demands on global companies for environmental responsibility, I believe that a strong case can be made that supports both a near term economic as well as carbon reduction; a two-fer.
Efficiency thru Cloud Engineering – closing on 0 [in-datacenter] distribution loss
While at Sun, working as CTO for Sun Grid in 2004-5, our power cost was over 10% of the cost of service (in our case $0.10 for each $1/cpuhr). Part of that cost was in the computers themselves, the other portion – often quite large, was the distribution of power within the data center from the ingress to the computers. Power Usage Effectiveness (PUE) was developed to measure the energy efficiency of a data center. PUE is basically the power provided to the data-center divided by the power used by servers/storage/network (useful work), a ratio. In those days the corporate datacenter were about 1.4-1.7, but today they are approaching 1.0 PUE.
WINNER = Public clouds due to new datacenter engineering and equipment
Just a month or so ago, Microsoft launched Project Natick, which sank a container datacenter off the coast of Scotland. Though they haven’t published their PUE, I can only imagine with free cooling and well insulated power lines that they might do pretty well [cooling = 0BTUhr] and with the accessibility to renewables – could operate at exceptionally low carbon footprint.
How Dirty Is that Energy Really (CO2/kWh) & Where is your Data Center?
Below are some great resources to help you calculate the CO2/KWH in your state.
For my home server (GPU rich) but efficient which runs at ~450 W & 24hrs = ~10.8kWh in VA.
How much carbon dioxide is produced per kilowatthour of U.S. electricity generation?
The U.S. Energy Information Administration (EIA) publishes average annual carbon dioxide (CO2) emissions factors for U.S. electricity generation in the State Electricity Profiles. Table 1 of each profile has the factor for the most recent year available. There are historical annual factors dating back to 1990 in Table 7. To find Table 7, see the link under Table 1 for: Full data tables 1-12. The factor is in pounds of CO2 per megawatthour (MWh). Divide the factor by 1,000 to convert the factor to pounds per kilowatthour (kWh).
You can calculate an average annual CO2 emissions factor for electricity generated in each state by type of fuel/energy source by dividing estimated CO2 emissions from electricity generation by type of fuel in each state by the amount of electricity generated by type of fuel in each state.
You can calculate the pounds of CO2 produced per kWh for specific types of electricity generators (prime movers) using different fuels by multiplying the heat rate of a generator (in British thermal units (Btu) per kWh generated), by the CO2 emission factor for the fuel (in pounds of CO2 per million Btu), and dividing the result by 1,000,000.”
One really interesting project that I have been watching is the Lefdal Mine in Norway. This converted mine has the ability to create “country trade zone” regions to deal with residency, and is substantially built in an environment – power, cooling, security and connectivity that is interesting – at least for archival storage, but potentially also for the use of data – AI, ML, HPC.
LMD is located near 350 MW renewable power production
PUE guaranteed at 1.15 – 24/7/365
CO2 emission is zero and the footprint limited + mine is built so construction impact limited
possibility of building out 120.000 m2 of whitespace in the facility
By any account, another interesting advancement in low impact data storage/usage.
If we can just concentrate and optimize workloads (decrease consumption) through resource sharing mechanisms, increase the utilization of a powered on server by 15% with a shift to cloud (increase concentration) and at the same time increase useful work from each kilowatt thru PUE gains toward 1.0 (increase work efficiency), we really can have a greener Internet and a cheaper IT estate¹. Of course, I have moved my home server into a container, on a public cloud [Azure] located in a state that supplies low CO2 power costs. I’m decreasing my costs and my footprint, you?
¹ note that we haven’t yet talked about the wide are network that cross-connects everything, it’s pretty power hungry itself! hint: WINNER = public clouds as the clouds are close to internet access points which are necessarily close to users!
UPDATE (01/2019): In case anyone wanted to know what that computer costs to operate. I was contacted by William @ ChooseEnergy.com who’s website provides economic data for energy cost per state here
Necessary cookies are absolutely essential for the website to function properly. This category only includes cookies that ensures basic functionalities and security features of the website. These cookies do not store any personal information.