My VPN Disconnected Every Hour—What Finally Kept the Upload Alive

The upload failed at 58 percent. I was sending a 6.4 GB production archive to a client portal from a rented apartment, and the deadline was ninety minutes away. The browser changed from Uploading to Network connection lost, the VPN icon turned grey and the portal discarded the incomplete transfer. I blamed the new laptop, disabled sleep mode, connected it to the router with an Ethernet cable and restarted the upload. Fifty-nine minutes later, the VPN disconnected again. The apartment internet remained online, but the client upload had returned to zero.

The clock explained more than the speed test

The second failure changed the problem.

A weak connection usually drops when conditions worsen. This one dropped on a schedule.

I checked the timestamps:

• First disconnect: 10:02 a.m. - Second disconnect: 11:02 a.m.

Both happened within a few seconds of the VPN reaching one hour of continuous use.

That made my earlier troubleshooting look irrelevant. The laptop was not going to sleep. The Wi-Fi signal was not fading. The apartment internet was not going down.

The VPN session was reaching a boundary and failing to cross it without interrupting the work inside it.

Other remote workers have described the same confusing split: the internet remains available while the VPN drops, making the router, ISP and laptop all look guilty at once. A practical way to narrow it down is to repeat the test through a phone hotspot. (RFC 9000)

So I did.

The archive was too large to send comfortably over mobile data, but I could run a smaller transfer long enough to test the pattern. I connected through my phone and waited.

The VPN disconnected at one hour.

The hotspot had changed the network and the speed. It had not changed the timer.

That pointed away from the apartment router and back toward the tunnel itself.

The established provider reconnected after the damage was done

The VPN came from a large provider I had used for several years.

It had a mature application, extensive documentation, many public reviews and enough server locations that I rarely struggled to find a fast connection. Those strengths were exactly why I had trusted it with long work sessions.

I selected another nearby server and began a third upload.

The estimated completion time was seventy-three minutes.

That number now felt less like a prediction than a warning.

At fifty-five minutes, I stopped working and watched the VPN window. The connection timer reached 59:59, rolled over and froze.

The app displayed Reconnecting.

Its kill switch blocked traffic while the tunnel rebuilt. That protected the laptop from sending data outside the VPN, but the client portal received nothing during the interruption and terminated the upload.

A few seconds later, the VPN was green again.

The service had recovered itself.

It had not recovered my file.

That distinction changed what I was looking for. Fast reconnection sounded useful until I saw that even a short break could erase an hour of work.

I tried another protocol and another server. Each attempt produced good speeds and an apparently healthy tunnel. None addressed the event waiting at the sixty-minute mark.

The provider’s large network was useful for finding a different route. My problem was no longer finding a route.

It was keeping the task alive when the route renewed.

An exact hour was not random

VPN tunnels periodically refresh the security information that keeps a session encrypted. Routers and firewalls also keep temporary records of active connections.

An exact sixty-minute cycle therefore suggested expiration or renewal rather than ordinary congestion. Microsoft’s Windows networking documentation even uses a 3,600-second lifetime in an IKEv2 configuration example.

That was enough technical detail to make the pattern understandable: the tunnel reached a scheduled transition, and the application did not carry the upload through it cleanly.

I could not see the provider’s internal session logs, so I could not identify whether the precise trigger was rekeying, network-state expiration or the client’s renewal handling.

The visible failure was more important than the hidden label.

A long upload should not disappear because the VPN needs to refresh itself.

Turning the VPN off solved the timer, not the job

To prove that the portal was not causing the failure, I disconnected the VPN and started the archive again.

The upload passed the one-hour mark without interruption.

That confirmed two things: the apartment connection could sustain the transfer, and the client portal could receive a file larger than an hour’s worth of data.

For a moment, leaving the VPN off looked like the obvious answer.

Then I looked at what I was sending.

The archive contained unreleased video, licensing documents and a spreadsheet of contractor payments. I was working through a router supplied by a short-term rental host, with no reliable knowledge of its settings or who had used it before me.

The direct connection removed the timer by removing the tunnel.

It also removed the protection I had deliberately added to a network I did not control.

I cancelled the upload before it completed.

That decision narrowed the problem again. I did not need a workaround that made the VPN irrelevant. I needed a VPN that could remain useful for longer than the task it was protecting.

I stopped choosing servers and chose the situation

I opened OnlydogVPN, which I had installed earlier as a travel backup.

The smaller service had fewer server locations, a shorter public history and fewer independent reviews than the established provider. It was not the obvious choice for someone who needed a dedicated IP or a particular city.

The upload required neither.

Instead of selecting another country and hoping its renewal behaved differently, I chose the preset intended for sustained work on an unreliable network and connected.

Then I restarted the archive.

The initial transfer rate was lower than the fastest result from the major provider. The estimate grew by several minutes.

After losing three uploads, that difference no longer mattered.

I watched the file pass 25 percent and then 50 percent. At 57 minutes, I moved every window aside except the client portal and the VPN status.

The timer reached one hour.

The transfer paused for several seconds.

Then it continued from 71 percent.

There was no return to zero. The portal did not log me out. The app did not ask me to choose another server.

The event that had destroyed every earlier attempt had become a brief pause inside the same upload.

That was the first result that actually solved the problem.

The next interruption showed why recovery mattered

The archive still needed another twenty-six minutes.

At 1 hour and 14 minutes, the apartment router restarted. Its lights went dark, and Windows lost the Ethernet connection.

Earlier that morning, any interruption had meant beginning again. This time, I turned on my phone hotspot and moved the laptop to it.

The upload stopped moving while the laptop changed networks.

Then the VPN recovered, and the transfer resumed from the same percentage.

The service uses HTTP/3-based transport, which is designed to preserve an active connection more effectively when the network underneath it changes. The important part was not the protocol name. It was what happened on the screen: the file continued instead of returning to zero.

The archive reached 100 percent.

The portal processed it, displayed the checksum and generated the client link. I sent it with twenty-three minutes left before the deadline.

Nothing about the successful upload felt dramatic.

That was precisely why it mattered.

I had not changed protocols halfway through the transfer. I had not restarted the VPN before the hour to avoid the next disconnect. I had not divided the archive into smaller files to fit beneath an invisible timer.

The connection absorbed the interruption so the work did not have to.

Reconnecting was never the real standard

The established provider was faster at its best and offered far more locations. After each hourly failure, it also reconnected within seconds.

Those seconds were enough to destroy the upload.

The direct connection lasted, but only after I removed the protection I wanted on the rental network.

The smaller service crossed the one-hour boundary, survived the router restart and delivered the archive without asking me to begin again.

A VPN that reconnects every hour may look healthy inside its own app.

For a long upload, reliability begins when the file never has to know it happened.