The first few days of 2018 have been busy for security professionals and IT admins. As Ars Technica put it: every modern processor has “unfixable” security flaws. There are fixes – sort of. But they come with a cost: computers will run up to 30% slower because of it, depending on the type of work being performed. A lot of the heavy lifting performed in large data centers is related to data processing in databases. Unfortunately, this is the type of operation that looks to have the worst impact of fixing the Meltdown and Specter vulnerabilities. A selection of tests shows that a performance reduction of about 20% is at least realistic (see details at phoronix.com). This is equivalent to increasing the power consumption of data centers by 25%. That is a lot of energy! With assumed growth in data center consumption of 5% per year we should expect the total electricity consumption by data centers globally to hit 480 TWh this year.
A 2013 estimate of the US use of data centers put the 2020 electricity use at 139 TWh, and the Independent reported that the 2015 global data center energy consumption was 416 TWh. For comparison, the total electricity generation globally is approximately 25000 TWh, so data center usage is not insignificant – and it is growing fast.
Our energy mix is still heavily dependent on fossil fuels, mainly coal and natural gas. Globally about 40% of our electricity generation is done based on coal, and another 30% is based on petroleum, primarily natural gas. In OECD countries coal use is on the decline but demand growth for electricity in particularly the BRICS economies still outnumbers the achievable growth in renewable energy generation in these areas. This means that short-term increases in global electrical energy generation will still be heavily influenced by fossil fuels – meaning coal and natural gas, and to some extent oil.
According to an article on curbing greenhouse gas emissions in the Washington Post, average coal fired power plants in the United States emits the equivalent of 1768 lb CO2/MWh, whereas for natural gas the average number would be around 800-850 lb/MWh. This corresponds to 800 g/kWh and 385 g/kWh for coal and natural gas, respectively.
Based on these numbers we can estimate an approximate value of extra CO2 emissions based on the Meltdown and Specter vulnerability fixes. If the unadjusted data center electricity consumption for 2018 is estimated at 480 TWh, and the bug fixes will lead to a 25% increase in consumption we are talking about and extra energy demand of 120 TWh for running our data centers. If 40% of that energy is generated by coal fired power plants, and 30% by natural gas, and the remaining by nuclear and renewable energy sources we are looking at 48 TWh extra to be produced from coal, and 36 TWh extra to be produced from natural gas. The combined expected “extra” CO2 emissions would be 52 million tons!
That is the same as all the climate gas emissions from the Norwegian economy in 2016 – including the entire petroleum sector (source: SSB). Another “yard stick” for how huge this number is: it corresponds to 1/3 of all emissions from U.S. aviation (source: EPA). Or – it would correspond to driving the largest version of a Hummer with a 6.2l V8 engine 3 million times around the earth (source: energy.eu).
This is a huge increase in emissions due to data processing because the CPU optimizations we have come to rely on since 1995 cannot be used safely. Also note that we have only included data centers in our estimate; this excludes all PC’s, Macs and smartphones that could see hits in performance too – meaning we’d have to charge our tech toys more often, thereby consuming even more electricity.