08 / LAB · NOTE Back to home

Six nodes for an island

Case study · simulation

September 27, 2026Barrel Digital

So far we have tested Barrel OS on a single Planet. A community is a different problem: more buildings, different day and night loads, and two ways to help each other. You can move energy, with a line that links the nodes. Or you can move compute work, which travels over fibre or radio and costs almost nothing.

We chose Lampedusa because it brings together, within a few kilometres, almost everything Barrel is designed for: healthcare, public administration, security, reception, school and water. It is a minor island not connected to the national grid: electricity comes from a diesel plant.

The case

NodeWhat it doesInstalled compute
Outpatient clinichealthcare, 24/7 service (critical)30 kW
Town halloffices and front desk, daytime30 kW
Reception centretranslation and transcription, more in summer20 kW
Coast guard and harboursecurity, 24/7 (critical)25 kW
Schooldaytime in school months; the rest of the time hosts deferrable jobs30 kW
Desalination plantproduces water when there is sun, with a two-day tank30 kW

Each node has one JA Planet (125 kW, 261 kWh) and 250 kWp of PV: 1.5 MWp and 1.57 MWh in total. The sun is real: hourly PVGIS series for Lampedusa, year 2023 (1,656 kWh per kWp). Building loads are hypothetical profiles we built, not measurements: we declare them in the code.

Five ways to run the same island

ScenarioEnergy from the island grid in one yearCompute demand servedWater demand produced with sun
Isolated nodes, without Barrel OS216 MWh100%64%
Microgrid, without Barrel OS89 MWh100%92%
Isolated nodes with Barrel OS42 MWh78%94%
Isolated nodes with Barrel OS and migrating compute47 MWhalmost 100%94%
Microgrid with Barrel OS17 MWh86%97%

When a node cannot manage on sun alone it draws energy from the island grid, i.e. diesel. Water the desalination plant cannot produce with sun is made by the existing plant: that energy is counted too.

Each scenario on two axes: diesel used in one year and share of compute demand served.Barrel OS v0.1 simulation · PVGIS 2023 sun · hypothetical loads

What we learned

  1. The problem is the night, not the sun. With 1.5 MWp the island produces twice what the six nodes consume. Without an orchestrator, though, batteries empty at night and 18% of consumption comes from diesel.
  2. Software matters more than the line. Barrel OS alone, without linking the nodes, brings diesel from 216 to 42 MWh. The microgrid alone, without an orchestrator, stops at 89.
  3. Moving compute is almost worth a line. If user requests can run on another node, practically all demand is served with 47 MWh of diesel, half the microgrid without software, and without laying a metre of cable.
  4. Together they do more. Microgrid and Barrel OS fall to 17 MWh, less than a tenth of the starting point.
  5. December remains the hard month. Even with the microgrid, in the darkest week batteries hit the bottom almost every night. With a second battery cabinet per Planet (configuration to confirm with the manufacturer) the microgrid with Barrel OS falls to 11 MWh and serves 99% of demand.
  6. Water is a second battery. The desalination plant runs when there is sun and the tank does the rest: with the microgrid 97% of water demand is produced with sun.
  7. In summer the sun is ahead. About half of annual production finds no use: there is room for more compute, more water or other island services.
Energy taken from the island grid, month by month.Barrel OS v0.1 simulation · PVGIS 2023 sun · hypothetical loads
The December week with least sun: six separate batteries on the left, one shared reserve on the right.Barrel OS v0.1 simulation · PVGIS 2023 sun · hypothetical loads
Above where compute demand originates, below where it actually runs: work goes to nodes that have energy, often the school.Barrel OS v0.1 simulation · PVGIS 2023 sun · hypothetical loads
Average desalination power by hour of day: with Barrel OS it concentrates in sunny hours.Barrel OS v0.1 simulation · PVGIS 2023 sun · hypothetical loads