Six Months Without a Cloud Drive

On 2026-01-08 I cancelled a cloud drive subscription I had held for nine years and replaced it with peer-to-peer file synchronization across four devices. The term was 180 days with a written rollback plan, same protocol as the password experiment. The term ended on 2026-07-07. Here are the numbers, including the ones that do not flatter the decision.

The setup

Syncthing, running on a desktop, a laptop, a NAS, and a phone. No server in the middle. Each device holds a full copy of the shared folders and negotiates changes directly with the others. Discovery over the local network where possible, relayed when not. 61.4 GB across roughly 118,000 files: documents, photos taken before the repatriation project, project archives, and the folder of scanned paper that I would genuinely grieve.

Setup time: 2.6 hours across all four devices, of which about an hour was me reading documentation carefully before touching anything, which I recommend.

The measurements

Metric	182 days
Data under sync	61.4 GB, 117,930 files
Change volume propagated	214 GB
Median propagation, same network	3.1 seconds
Median propagation, remote device	46 seconds
Conflict files created	37
Files lost	0
Manual interventions required	9
Monthly cost	about $1.40 (NAS electricity share)
Previous subscription	$9.99

The saving is real but small enough that I would not run this for the money. $8.59 a month is not a reason to change anything. The reason is further down.

Failure mode one: conflicts are your problem now

A cloud drive has a server, and a server has an opinion. When two devices disagree, the server's copy is the truth and you move on. Remove the server and you have removed the referee. What you get instead is a .sync-conflict file sitting politely next to the original, waiting for a human to decide.

Thirty-seven of them in six months, every one from the same behavior: editing a file on two devices without letting the first edit propagate. None cost me data, because the conflict file preserves both versions, which is the correct design. All of them cost me attention. Call it 4 minutes each, so 25 minutes a month of being the referee. That is the honest tax and nobody puts it on the comparison chart.

Failure mode two: the phone is a tourist, not a peer

This is the one I did not predict. Peer-to-peer sync assumes peers, and a modern mobile operating system does not permit an application to behave like one. Background execution gets suspended aggressively, radios sleep, and the sync process wakes when the OS feels generous.

Median staleness of the phone's copy, sampled daily: 4.6 hours. Worst observed: 31 hours. On the three desktop-class devices, median staleness was under 10 seconds.

So the four-device mesh is really a three-device mesh with a device that checks in. For photos and reference documents this is fine. For anything I need current on the phone, it is not, and I stopped pretending otherwise around week five. The workaround is unglamorous: I keep a small set of genuinely mobile documents in a folder I sync deliberately, and I accept that the phone is a cache with opinions rather than a full participant.

What I actually bought

Not savings, and not uptime. The thing I bought is that there is no longer a party who can lock the folder.

That sounds abstract until you price it. A provider can suspend an account for a violation determined by a classifier, with an appeal measured in weeks, and during those weeks the files are gone from your reach. The probability in any given year is low, under half a percent by my estimate, and I will not pretend to more precision. But the impact is total and the recovery is out of your hands, which is the shape of risk people systematically underprice.

Now the failure modes are mine. The disk dies, I restore. The sync misbehaves, I fix it. Both are problems with a defined owner and a known procedure, and I prefer those to problems that arrive as an email saying my account has been reviewed.

The caveat that matters most

Sync is not backup. This deserves its own line because getting it wrong is how people lose everything. If I delete a file on the desktop, that deletion propagates to all four devices in seconds, faithfully and irreversibly. A replicated copy of a mistake is still a mistake, in four places, very quickly.

So underneath the sync layer there is a separate versioned backup running to two destinations on a schedule the sync system cannot touch. Those are different systems solving different problems and conflating them is the single most common self-hosting error I see.

I am about to write more about that, because my most recent restore drill did not go the way I expected.

Verdict

Rollback plan deleted. Confidence 0.88 that this is still running in 18 months, with the residual mostly covering the scenario where the phone situation degrades further and I decide the topology is not worth the referee work.

For $8.59 a month saved and 25 minutes a month spent, my files stopped having a landlord. Run your own numbers. The 25 minutes is the part people forget to count.