Tea Time – 11/29/2015 – The Elastic Internet

It is interesting that Google alone boasts more than 2.4 million servers (http://www.circleid.com/posts/20101021_googles_spending_spree_24_million_servers_and_counting/) and Microsoft had crossed the 1 million server mark back in 2009 (http://www.extremetech.com/extreme/161772-microsoft-now-has-one-million-servers-less-than-google-but-more-than-amazon-says-ballmer).  Assuming that each server draws about 750 watts (between compute and storage nodes) that means at least 3.4 million servers times 750 watts or 2,550,000 kilowatts of power consumption.  On top of this, there is a requirement for cooling.  Suppose that the cooling is done via heat-pumps (air conditioners).  With an energy efficiency ratio (EER) of 8 (https://www.e-education.psu.edu/egee102/node/2106) and a conversion of 3.41 BTU/hr per watt  means an additional 1,086,937 kilo watts must be included to remove the heat of the servers.  So, 3,636,937 kilowatts total to keep the servers online.  Coal energy is measured in kilowatt hours per ton (https://www.eia.gov/tools/faqs/faq.cfm?id=667&t=2) so we convert to kwh by multiplying by the number of hours per year -> 365.25 * 24 = 8,766 hours / year.

From the above calculations, the total KW per year needed for Google and Microsoft is 31,881,394,125 kwh.  Using a conversion of 1904 KWH per ton of coal, it thus takes 16,744,430 tons or 15,222,209,091 kilograms of coal per year to keep them running.  The density of coal depends on the type.  However, we can use an average between 641 and 929 kg / cubic meter (http://www.ask.com/science/bulk-density-coal-e55167b75b4deafc) or 785 kg / cubic meter to figure out the size of the coal lump we will need.    Dividing the number of kg of coal required by the density gives us 19,391,349 cubic meters.  Taking the cube root of that gives us a cubic lump about 269 meters per side.  For those using American units, we multiply by 3.28  feet / meter to arrive at 881 feet per side or a little less than 2 and a half American football fields per side.

This is where elasticity comes into play.  An elastic server application uses software driven automation to determine whether a server is heavily loaded or lightly loaded and can take steps to remove power from lightly loaded servers. Thus, depending on how loaded the servers are at a given time of day, the number of servers drawing power may be decreased and thus the size of that lump of coal.  A really good application for such automation is streaming video. The majority of the population sleeps at night and works during the day.  Thus, servers that provide video to populations can be assumed to have their heaviest load during the evening hours when people are off of work but not asleep.  The rest of the day would be a light load on those servers.  Assuming 8 hours of heavy load out of a day this means a savings of 2/3 or 10,148,394 kg of coal per year. Now wouldn’t that be fabulous?