Impact

Energy

How much electricity a Drupflare site uses against the hosting it replaces, and where the saving comes from. Energy Per View and Fleet Energy extend the model; Environment turns the totals into carbon and water, and Economics prices them. Every figure carries a label defined on Impact. Numbered citations such as [1] refer to the References list on Impact.

Pages

Energy Savings

Total energy of one site over a year, idle included, for drupflare and for a production deployment. A conventional deployment draws power whether or not anyone visits; drupflare's figure is the CPU time its views use, logged-in views included. Every cell is derived (modelled): no source sizes a production node, so the production shapes are modelled, and drupflare's energy is derived from measured CPU time on the traffic mix. A month is a twelfth of each figure. Generated by bun scripts/economics/perview.ts.

This is the sleeping state, a site that holds nothing between requests. A site kept warm, or one that wakes cold, draws more. Operating States prices both, and the saving against production stays above 99.5% in every state.

views a month drupflare, energy a year production, energy a year energy saved a year saving against production saving against peak-sized
10,000 0.525 Wh 224 kWh 224 kWh 426,766x, 99.9998% 742,503x, 99.9999%
100,000 2.14 Wh 224 kWh 224 kWh 104,856x, 99.9990% 182,432x, 99.9995%
1,000,000 18.3 Wh 224 kWh 224 kWh 12,274x, 99.9919% 21,355x, 99.9953%
10,000,000 180 Wh 224 kWh 224 kWh 1,249x, 99.9199% 2,173x, 99.9540%
30,000,000 538 Wh 224 kWh 224 kWh 417x, 99.7601% 725x, 99.8621%

Production is three regions with a high-availability web pair and database pair in each, sized at 15% utilisation. Peak-sized is the same three regions with 8-vCPU nodes at 10%, 390 kWh a year. The multiple is the ratio of the two totals. It grows as traffic falls because drupflare's total has no idle term, so quote the percentage.

What those savings compare to, at 10,000,000 views per site per month, derived from the same totals by bun scripts/economics/comparison.ts --md:

sites against electricity carbon servers not built
1 production one US home for 7 days 180 lb CO2e 1/14 of one server not built
1 peak-sized production one US home for 13 days 330 lb CO2e a seventh of one server not built
10 production one US home for 75 days 1,800 lb CO2e a half of one server not built
10 peak-sized production one US home for 131 days 3,300 lb CO2e 1 server not built
100 production 2 US homes for a year 1 car off the road 7 servers not built
100 peak-sized production 3 US homes for a year 3 cars off the road 15 servers not built
1,000 production 20 US homes for a year 18 cars off the road 75 servers not built
1,000 peak-sized production 36 US homes for a year 32 cars off the road 150 servers not built

Every figure is rounded down, so each is a floor. A server is the share of one 640-thread host the model assigns to the deployment: 12 nodes at 4 or 8 vCPU each, counted in whole servers and, below one, as the largest unit fraction that fits, derived (modelled). The host is the SPECpower result for a Dell PowerEdge R6725 with two AMD EPYC 9845 chips, 320 cores and 640 hardware threads, which idles at 135 W and draws 711 W at 100% load [1]. A US home is the EIA's 2022 average of 10,791 kWh a year [2], and a passenger vehicle emits about 4.6 t of CO2 a year [3].