One of the arguments for [Woodchuck][http://hssl.cs.jhu.edu/~neal/woodchuck] is that it can save energy. In this post, I want to examine that claim a bit more quantitatively.

To determine whether or not Woodchuck can save energy, we first need to know approximately how much energy the activities we are interested in consume. To measure this, I charged my N900 until the battery was full, then I started some activity and let it run until the device turned off. Every five minutes, I queried the battery's state (voltage, mAh and whether the device was being charged) and wrote it to an SQLite database. The activities that I measured were: streaming or playing an mp3 file at various encodings, downloading over WiFi at different speeds, having the LCD on, and idling. Some of the results are summarized in the table below. Keep in mind that a full charge has approximately 18 kWs (= 5 Wh).

Data Acquisition Activity Watts Energy Consumed Relative to Idle
3G Play 56 Kb/s stream 1.00 12.5
Edge Play 56 Kb/s stream 0.96 12.0
WiFi Play 56 Kb/s stream 0.75 9.3
Flash Play 56 Kb/s files 0.28 3.5
Flash Play 128 Kb/s files 0.27 3.4
Flash Play 320 Kb/s files 0.32 4.0
WiFi Download at 4.7 Mb/s 1.23 15.4
WiFi Download at 1.0 Mb/s 0.91 11.4
WiFi Download at 256 Kb/s 0.76 9.5
None Idle, LCD on 0.27 3.4
None Idle 0.08 1

The first thing to notice is that streaming over a network connection is expensive: streaming over 3G consumes 20% of the N900's battery capacity per hour. Although it is possible to save a bit of energy by using Edge or WiFi, the improvement is marginal. Playing back audio data saved on flash requires significantly less energy---just 30% as much. In other words, if all you do is use your N900 to listen to audio, listening to audio data saved on flash will allow you to listen to more than 3 times as much audio on a single battery charge than if you were to stream that data.

It is not always possible to ensure that the data is saved on flash. In this case, the best approach is to download the data as fast as possible: although downloading over WiFi at 4.7 Mb/s (the maximum sustainable throughput I observed) requires more energy than downloading at, say, 256 KB/s, the required energy per bit is significantly lower.

To put these values in perspective, I measured how much energy the system consumes at idle and with the LCD on. I think it is not surprising that having the LCD on consumes significantly more power than not, however, I was surprised that the network uses 3 times as much energy as having the LCD on.

What do these values mean for Woodchuck? Woodchuck tries to schedule downloads to occur when conditions are good. In terms of energy, conditions are best when the device is connected to the mains. I charge my N900 about every two days. Only updating my subscriptions every two days is not often enough: I don't want the news from a day and a half ago; many blogs that I read are updated daily; and, my calendaring information should be synchronized constantly. In this case, fetching the data as fast as possible over WiFi when the signal is strong is the next best approach.

To understand the possible savings, consider the case where 8 hours of audio, about 200~MB of data, are prefetched over WiFi. At 4.7~MB/s, this requires 420~Ws (2.5% of the battery's capacity). If a user listens to 30 minutes of audio (25~MB) on the commute home, only an additional 480~Ws (2.7% of the battery's capacity) are required. Streaming 30 minutes of audio over 3G requires 1800~Ws, twice the amount of energy to prefetch 8 times the data and listen to the same audio. Thus, even with a cache hit rate of 12%, prefetching uses just half of the energy needed to stream.