On January 7, 2010, someone here installed a BOINC client and joined World Community Grid under the name sophron.io. Nobody has turned it off since.

That is most of the story. What follows is the accounting, which turned out to be more interesting than the decision was.

What the thing actually does

World Community Grid is a volunteer computing platform, hosted today by the Krembil Research Institute in Toronto. Researchers bring it workloads that decompose cleanly into small independent pieces — docking a few thousand candidate compounds against a protein target, scoring a slice of a genomic dataset, running one cell of a climate grid. The platform chops the job into work units and hands them out to whoever has a machine sitting idle.

Your machine takes a unit, computes it, sends the result back, takes another. Units go to more than one volunteer and the answers get compared, because a network made of strangers’ desktops is not a trustworthy substrate and the platform is designed on that assumption.

It is an old idea. SETI@home made it famous in 1999, and it still works for one narrow but real class of problem: embarrassingly parallel, light on data transfer, heavy on arithmetic, and not so sensitive that it needs a controlled environment.

The accounting

Sixteen years in, under one team name:

• 30 years, 207 days of cumulative CPU run time
• 116,140 work units returned
• 66.5 million points
• 70 badges across 17 research projects

The run time is the number worth pausing on. Thirty years of processor time accumulated over sixteen calendar years means an average of roughly two cores running continuously, forever — which is a fair description of what actually happened. Not a cluster. Not a donation. Just machines that were already bought, already racked, already drawing power, doing something in the gaps instead of nothing.

A list that reads like a timeline

The badge list is an unintentional record of what the world was worried about, year by year:

FightAIDS@Home and Human Proteome Folding. Help Conquer Cancer, Help Fight Childhood Cancer, Help Cure Muscular Dystrophy. Computing for Clean Water. Drug Search for Leishmaniasis. GO Fight Against Malaria. Say No to Schistosoma. Outsmart Ebola Together — that one dates itself. OpenZika — so does that. Smash Childhood Cancer. Microbiome Immunity Project. OpenPandemics – COVID-19. And running underneath nearly all of it, Mapping Cancer Markers, which alone has taken twenty years of our processor time.

Today the cycles go to Mapping Arthritis Markers, the Africa Rainfall Project, and Mapping Cancer Markers. The rainfall work is the odd one out and the one we find most interesting: modelling precipitation over sub-Saharan Africa at a resolution fine enough to be useful to a farmer deciding when to plant, which is a computation nobody was going to buy a supercomputer for.

The engineering notes, such as they are

This is not a hard thing to run, and most of what there is to say is about restraint.

The hardware is already sunk. A server provisioned for peak load is idle most of the day. A workstation is idle most of the night. That capacity does not bank — an unused core hour is gone. Volunteer computing is one of the few ways to convert it into anything at all.

Power is the real cost, and it is not zero. A machine at full utilization draws meaningfully more than an idle one, and it makes more heat, which the cooling then has to move. This is cheap, not free, and anyone claiming otherwise has not read a power bill. Budget for it honestly.

Run it on hardware you own. Metered cloud compute is the wrong place for this. You would be converting someone’s donated science into a line item on your own invoice at retail rates. Physical machines you already pay for, on power you already pay for — that is where the arithmetic works.

Let it lose. The client is built to yield: cap the core count, set it to back off under load, let it run at low priority. Ours has never been the reason a real job was slow, because it is configured to be the first thing to give up its slice. If volunteer computing ever competes with production work, it has been misconfigured.

The honest part

Donated cycles do not substitute for research funding. They are useful for one shape of problem and irrelevant to everything else, and a project that receives them still needs people, labs, and grants. The points and badges are a leaderboard, which is to say a game — a well-designed one that has kept volunteers engaged for two decades, but a game.

The platform is also not immune to the world. World Community Grid went dark for most of 2022 while it moved from IBM to Krembil, and volunteers had nothing to do but wait. If you sign up, sign up for the version of this where it sometimes stops.

None of that changes the arithmetic. The machines are running anyway.

If you want to do this

Install the BOINC client, pick World Community Grid, and choose your projects. Set a core limit you will not notice. Join a team if you want the numbers to add up somewhere visible, or don’t. The whole setup takes about ten minutes, and then it is a thing you have done rather than a thing you keep doing.

Ours has been running since a Thursday in January 2010. It is still running now.