Tuesday 14 April 2009

Spam and its Carbon Footprint

All uses of the Internet have an impact on climate change. Sadly, that includes the less-savory uses.

Spammers dumped 60 trillion messages onto the Internet in 2008. As the climate-change consensus becomes overwhelming, it's high time we looked at the environmental impact of spam.

Recently, McAfee commissioned climate-change consultants ICF calculate the carbon footprint of spam. McAfee also asked me to help. We calculated the energy use associated with each stage in the lifecycle of spam, including the energy used to transmit, process, and filter spam.

Globally, the annual spam energy use is 33 billion kilowatt-hours, or 33 TWh—that's as much electricity as 2.4 million U.S. homes use, with the same greenhouse gas emissions as 3.1 million passenger cars using 2 billion U.S. gallons of gasoline.[1]

Two Surprising Conclusions

Far from being a net consumer of energy, spam filtering actually saves an incredible amount of energy. Imagine if all the spam filters in the world were switched off for a day. It would actually increase the carbon footprint of spam by at least five times.[2] In other words, spam filtering saves 135 TWh of electricity per year—that's like taking 13 million cars off the road.

But we could do even better. Imagine if every inbox were protected by a state-of-the-art spam filter. We could save about 75% of the spam energy used today—25 TWh per year;[3] that's like taking 2.3 million cars off the road.

Other Results

The average greenhouse gas emission associated with a single spam message is 0.3 grams of CO2. That's like driving 3 feet (1 metre), but because of the annual volume of spam, it's like driving around the Earth 1.6 million times.[4]

A year's email at a typical medium-sized business uses 50,000 KWh, more than one fifth of which is associated with spam.[5]

Filtering spam is all well and good, but fighting spam at the source can have even better results. Taking McColo offline in late 2008 saved energy equivalent to taking 2.2 million cars off the road, before spammers rebuilt their sending capacity.

Energy use associated with spam is mainly consumed by end-users deleting spam and searching for legitimate email ("false positives"). Only 16% of energy use is from spam filtering itself.

Notes

My role in the McAfee project was to help ICF build a model that accurately reflected where energy was used in producing, transmitting, filtering, and dealing with spam. To this end I provided consultancy and data, plus some analysis of the results.

The data came from my 25 years of experience with email and spam, cross-correlated with data from other researchers (including McAfee and McAfee's competitors).
  1. 33 TWh of electricity use emits 17 million tonnes (19 million U.S. tons) of CO2, equivalent to 3.1 million passenger cars, burning 7.6 billion litres of gasoline (2 billion U.S. gallons), or 2.4 million U.S. homes' electrical usage.
  2. Switching off spam filtering for one day would multiply spam in the average inbox by 5x and multiply false positives by at least 10x. While no energy would be used by spam filters, this reduction is vastly outweighed by the energy used by end-users coping with spam.
  3. Most inboxes are protected by spam filters, but many of them are less accurate than the best filters. Some inboxes are still completely unprotected. State-of-the-art filters can achieve better than 98% effectiveness/0.01% false positives and use less power: assumes 25% power saving over legacy spam filters.
  4. Based on passenger car averaging 20 miles per U.S. gallon; mean equatorial circumference 40,041 km (24,870 miles).
  5. Refers to an organization with 200 average business email users.

Incremental Energy

In any calculation such as this, there's always the concern that we're double-counting energy that would have been used whether or not there was spam. Let me assure you that this isn't the case. I only wanted to be involved in the project if we were measuring this meaningfully.

So the data and calculations were carefully designed so as to only measure energy that is used as a direct result of there being spam. In other words, it is "incremental" energy.

PCs and servers use less energy when idle than when doing "work"—in most cases it's this additional energy that we measured.

More About the Methodology

Some wags have complained that ICF doesn't publish its methodology. They have clearly not read the full report, including the appendix, which helpfully titled, err, Statement of Methodology. Perhaps they're confused by the 8 page summary?

Download Full Report

You can now get the full 28 page version at the usual place.

Radio Interview

ORF interviewed me on Friday. Download it here: richi-on-orf.mp3.

7 comments:

Anonymous said...

I disagree: Turning off spam filters would not multiply false positives by a factor of 10, but rather reduce them to 0 (since there is not filtering).

Richi Jennings said...

Sadly, that's not true.

Studies have shown that people are much worse than spam filters, where false positives are concerned. In essence, people get tired quickly.

Sorry, I don't have a cite to hand, but a 10x difference is a conservative estimate.

Joe Wagner said...

Could you please expand or cite (a viewable citation) where your energy numbers come from? The report claims 62,000,000,000,000 spam per year and total energy used for spam as 33,000,000,000 KWh per year (http://img.en25.com/Web/McAfee/CarbonFootprint_12pg_web_REV_NA.pdf")

That seems to work out to 0.000532258 kilo-watt hours per spam or a lovely half a watt-hour per spam. A Segway supposedly travels 24 miles on a single 52 watt-hour charge (or 100 spam?). My humble laptop may have to deal with thousands of spam on one battery charge, and I'd bet my laptop's considerably smaller that the battery packs of a Segway...

So there seems to be some fundamental presumptions being built into the calculations aside from raw electricity being used by any manner I can think of off hand. Can you shed any light?

Thanks.

Lars said...

Does the study only take into account the extra energy used by email readers' client machines? Or does it also consider the extra energy used by email relay servers to relay spam, and/or the energy used by hijacked machines to originate spam?

Richi Jennings said...

Joe, there's a variety of places where energy is used as a result of spam and a variety of spam filtering scenarios.

It's quite a bit more complex than the picture you paint.

The full 25 page report goes into detail about how the figures were calculated. The 8 page summary you link to is a bit more limited!

Richi Jennings said...

Lars, yes all of the above and more. I recommend reading ICF's full report.

Anonymous said...

Sorry for my bad english. Thank you so much for your good post. Your post helped me in my college assignment, If you can provide me more details please email me.

Post a Comment