Continuous protection: changes saved within minutes
A continuous job saves each change within minutes of making it. Every save is a point you can go back to, and old points thin out by themselves.
On this page
A scheduled job looks at all your files at set times. A continuous job is told by Windows what changed and saves just that, within minutes. Every save is a point: the folder as it was at that moment, which you can open, restore or put back. Continuous protection is part of Pro and Server, their trials included.
New to it? Protect your files continuously, step by step sets a job up in about five minutes.
Wizard Schedule step - Continuously
Turning it on#
- Create a job, or open one and click Edit.
- On the When should it run? step choose Continuously.
- Set the rhythm and the settings below, then save.
The first save is a full backup, as with any job. From then on each point holds only the files that changed. A continuous job is always incremental: a mirror keeps one copy only, so the wizard asks for another type.
On Home and Free, Continuously is greyed out with a note. A continuous job saves changes within minutes only while the PC runs Pro or Server, or their trial. When the trial ends on Home, a continuous job made during it is saved every 15 minutes instead (every hour on Free), without the ransomware guard, and says so: Saved every 15 min · continuous needs Pro. Its settings stay for when Pro is back. Genie, asked for a continuous job on Home, drafts one that runs every 15 minutes.
The settings#
| Setting | Choices | What it does |
|---|---|---|
| Save changes at most every | 1, 5, 15 or 30 minutes (5 by default) | How long a change waits before it is saved, at most. A quiet PC saves nothing. |
| Files in use, like Outlook's mail, at most every | 15, 30 or 60 minutes (30 by default) | Files a program keeps open are read from a snapshot, at most this often, and as soon as their program closes them. Needs service mode (see below). |
| Start a fresh full backup every | 7 to 365 days (30 by default) | A new full starts a new chain, so restores never depend on a very long one. |
| Keep versions for | 1 to 3650 days, or empty | How long points are kept. Empty keeps them while there is space. |
| When space runs low, remove the oldest versions first | on by default | The space guard (below). |
| Watch for ransomware | on by default | See The ransomware guard. |
| Keep versions on this PC while the destination is away | off by default | Up to 5 GB of points kept on this PC until the destination is back, always leaving the drive room (below). |
The conditions under it apply too: on battery and in quiet hours, saving waits until they no longer hold. See Schedule conditions.
Files in use#
Some files stay open all day: Outlook's mailbox (.pst), a database, a virtual disk. Without service mode they are saved when their program closes them; the Schedule step then says so, with Turn on service mode to save them while open. With service mode on, the agent reads them from a Volume Shadow Copy snapshot at the rhythm you chose, without closing anything. A new file its program still holds while writing it (a video being copied in, an export) waits until the program lets go of it, so no point keeps half of it; one still held after that rhythm, like a new mailbox, is read from a snapshot then. See Running backups with nobody signed in. Turning service mode on or off keeps each continuous job as it is: its points, a hold of the ransomware guard and the points kept on this PC go across with it, and it goes on from its last point.
Large files#
A file of 16 MB or more (a mailbox, a disk image, a video project) is saved in blocks: each point stores only the parts of it that changed, not the whole file again. It waits until it has not changed for 2 minutes, or until its program closes it, so a file still being copied or saved goes in whole; the points meanwhile keep its earlier version. One of 1 GB or more is read in full each time, so it is saved at most as often as files in use, and at once when its program closes it. See Backing up Outlook mailboxes and other large files for what that saves with a real mailbox.
How changes are found#
| Where the files are | How | What it means |
|---|---|---|
| An NTFS or ReFS drive | Windows change journal | Windows' own list of changes. Nothing is read until something changes. |
| A drive without a journal (exFAT, FAT32) | Folder watch | Windows tells the agent of each change while it runs. |
| A network share | Checked every 15 minutes | A share sends no reliable notice, so its folders are looked at. |
The job page shows which one each job uses, under How changes are found.
The first time a job uses a drive whose journal is off, or too small to keep the changes made while GBM AI was closed, the app asks once: Turn on the Windows change journal on C:? or Give the change journal on C: more room? Windows asks for permission. Not now works too: a drive whose journal is off is followed by a folder watch, and one whose journal is small keeps using it; either way the job looks through all its folders whenever the changes could not be followed, after a restart for example.
A file written in the last 30 seconds that its program has not closed yet waits for the next point: it may still be being written. On a drive followed by a folder watch, every file written in the last 30 seconds waits, since the watch cannot see a program close it. The job's line says 2 files are still being written.
Watching it work#
The job's line on Home and on its page says what it is doing:
- Protected: everything up to Last saved 14:31 is in a point.
- Saving changes: a point is being saved now. A point takes seconds.
- Waiting, with why: Waiting while on battery, Waiting for the backup to finish, Making the first full backup, Waiting for its destination, Waiting while an update installs.
- Paused: scheduled backups are paused (notification-area icon ▸ Pause scheduled backups). A job the ransomware guard holds says Paused: changes look like ransomware. After the trial, until you choose an edition, it says Paused until you choose an edition.
- Not saving changes: something went wrong; the line says what.
The line also shows the job's rhythm and what is still to come: Changes in 3 folders, 1 file in use is saved when its program closes it.
Activity groups the points by the hour, for example Saved changes 12 times: 48 files, 3.2 MB, and so does the last-run strip on the page of a job with a second copy (Saved changes 12 times · 3.2 MB). A point that failed has its own line. The notification-area icon says Continuous protection on or paused, names a job the ransomware guard holds, and says when a job is saved every 15 minutes because continuous protection needs Pro.
Run now on a continuous job (Back up now in the notification-area menu) saves a point now, even while scheduled backups are paused, on battery or in quiet hours. It is refused, with the reason, while the ransomware guard holds the job, while backups are paused until you choose an edition, while its destination is away with nothing kept on this PC, or while its source is away. While the guard holds the job, Run as full is off too.
GBM AI does not tell you of every point. It tells you when protection stops: three points in a row failing (at most once an hour), or the destination away for a day.
How versions thin out#
Points pile up fast, so older ones are merged by themselves, about once an hour:
- every point from the last day;
- one an hour for the last week;
- one a day for the last month;
- then one a week, until Keep versions for ends.
A merged point holds the newest version of each file from the points it replaces. The timeline shows it as 12 saves in one.
The space guard#
When the destination has less than 5% free, or not enough for the next point, the space guard makes room before saving: it merges early, then removes the oldest points. A file's newest version always stays, the last version of a deleted file stays at least 30 days, and the points of the last day are never merged. When nothing more can go (a small drive that fills within a day, or the guard off), a point that does not fit fails: the job says the destination is full, and after three you are told. What did not fit is saved as soon as there is room again.
Where the points go#
- A USB or internal drive: at the rhythm you chose.
- A network share or an S3 bucket: at most every 15 minutes, whatever the rhythm. Each point of a bucket job is whole in the bucket when it is saved, and with Object Lock each point is locked from its start. See Locking backups in an S3 bucket.
- A second copy (a drive, a share or a bucket) gets the new points about every 15 minutes, once each is checked: the copy is looked at every 5 minutes and made at most every 15. The job's page says when: 3 points to copy at 14:37. Points merged on the backup are merged on the copy too. See A second copy.
When the destination goes away#
Unplug the drive, take the laptop out of the office, or lose the connection to the bucket, and the job simply waits: Waiting for its destination. Not connected since 09:12. Changes are kept and saved when it is back. Nothing fails and nothing alerts. When the destination is back, one point takes everything that changed meanwhile. A USB drive that comes back under another letter is found by the drive itself, as for a scheduled backup. After a whole day away, GBM AI tells you once.
With Keep versions on this PC while the destination is away on, the job keeps saving points on this PC instead, at its own rhythm and encrypted as the job is. The line says Saving on this PC and 3 points are kept on this PC (240 MB) until the destination is back. Files of 16 MB and more wait for the destination, as a point of one would hold all of it. When the space (5 GB) is full, or the drive would keep less than its margin free (a twentieth of the drive, between 2 and 10 GB), the oldest points kept here are merged; when nothing more can merge, newer changes wait for the destination.
When the destination is back, the points kept here go to it first, oldest first, are checked there, and leave this PC. Until then, Restore lists them as On this PC and restores from them like any other point.
When the source goes away#
A job of a folder on a NAS or a USB drive waits the same way when its source is not there: Waiting for its source. Source not connected since 09:12. What changed there is saved once it is back. Nothing is recorded as deleted and nothing fails. When the source answers again, the next point saves what changed there.