Every hosting provider worth paying backs up your virtual machine, and MassiveGRID takes a daily backup of every managed server with snapshots on demand. That is the right protection against a lost disk or a ruined operating system, and the wrong protection against the two things that actually destroy databases: a bad deployment that deletes rows at 14:07 and a corruption that has sat quietly inside a data file for three weeks. Both need SQL Server's own backups, taken on SQL Server's own schedule, verified and restored regularly. This guide sets that up on a hosted Windows server: which recovery model to use, what full, differential and log backups buy you, how to schedule them with a maintenance plan or Ola Hallengren's scripts, how to send them off the server with BACKUP TO URL, how long to keep them, and how to prove once a month that they restore. It applies equally to a Windows VPS you run yourself and to the managed SQL Server hosting where MassiveGRID runs the operating system and you own the databases.

Two kinds of backup, and why you need both

A VM backup copies the whole virtual disk. It restores a server to the state it was in at the moment of the backup: operating system, SQL Server binaries, data files, everything. It is taken outside SQL Server, usually once a day, and it is what rescues you from a failed update, ransomware on the OS or an administrator deleting the wrong folder. On MassiveGRID's managed servers it runs daily with on-demand snapshots, and on the HA cluster it is complemented by three-way replicated storage, so a failed disk is never a restore event at all.

A database backup is taken by SQL Server, is transactionally consistent, and, with log backups, can restore to any second you choose. It is the only backup that gives you point-in-time recovery, the only one you can restore to a different server to pull out one table, and the only one that runs CHECKSUM over every page while it reads them. A daily VM backup leaves up to 24 hours of exposure; a log backup every 15 minutes leaves 15.

The provider's backup protects the server; your backup protects the data. Run both, and keep at least one copy of the database backups somewhere other than the server they were taken on.

Recovery models and the RPO question

Before any schedule, answer one question per database: how much work can we afford to lose? That is the recovery point objective (RPO), and it decides the recovery model.

Recovery modelLog behaviourBest achievable RPOUse it for
SimpleLog space is reused automatically; no log backups possibleTime since the last full or differential backup (hours)Development, staging, reporting copies, data warehouses rebuilt from source
FullLog grows until a log backup is taken; point-in-time restoreThe log backup interval (minutes)Anything whose rows cannot be re-created: orders, bookings, ledgers, CRM
Bulk-loggedLike Full, except bulk operations are minimally logged and not point-in-time restorableLog interval, with gaps around bulk loadsTemporarily, during a large import

The classic mistake is a production database in the Full model with no log backups: the log file grows until the disk is full, and the server stops. If you cannot or will not take log backups, use Simple and accept the longer RPO; do not leave Full unattended. Check what you have:

SELECT name, recovery_model_desc, log_reuse_wait_desc FROM sys.databases WHERE database_id > 4;

A log_reuse_wait_desc of LOG_BACKUP on a Full-model database tells you the log is waiting for exactly the backup nobody scheduled.

Full, differential and log: the schedule

The three backup types nest. A full backup copies every page in use. A differential copies every page changed since the last full, so it grows through the week and is restored on top of the full. A log backup copies the transaction log since the previous log backup and is restored in sequence on top of the last full or differential. A restore to 14:06 on Thursday is: Sunday's full, Wednesday night's differential, then every log backup from midnight up to and including the one that spans 14:06, with STOPAT.

A schedule that fits most single-server applications:

Always use WITH CHECKSUM, which makes the backup verify each page's checksum as it reads and writes it, and COMPRESSION, which cuts size and I/O by half or more on typical data (available in every edition since 2016 SP1, including Express for full backups; Express cannot run SQL Server Agent, so schedule with Task Scheduler and sqlcmd). The plain T-SQL, so you can see what any tool is generating for you:

BACKUP DATABASE shop TO DISK = 'D:\SQLBackup\shop_FULL.bak' WITH COMPRESSION, CHECKSUM, INIT;
BACKUP DATABASE shop TO DISK = 'D:\SQLBackup\shop_DIFF.bak' WITH DIFFERENTIAL, COMPRESSION, CHECKSUM, INIT;
BACKUP LOG shop TO DISK = 'D:\SQLBackup\shop_LOG.trn' WITH COMPRESSION, CHECKSUM, INIT;
RESTORE VERIFYONLY FROM DISK = 'D:\SQLBackup\shop_FULL.bak' WITH CHECKSUM;

Write backups to a volume other than the one holding the data files where the server has one, and never to the C: drive, where a full disk takes Windows down with it.

Maintenance plans or Ola Hallengren

Two ways to schedule all of this on an instance with SQL Server Agent.

Maintenance Plans are the wizard in SSMS under Management. They produce SSIS packages that run as Agent jobs, with a Back Up Database task per backup type, a Check Database Integrity task and a Maintenance Cleanup task for old files. They are fine for a single server with a few databases, and they are the right answer when the person maintaining the server is not a DBA. Their weaknesses: new databases are not picked up unless you chose "All user databases", the cleanup task is easy to misconfigure, and index maintenance is all-or-nothing.

Ola Hallengren's Maintenance Solution (free, MIT-licensed, from ola.hallengren.com) is one script that installs stored procedures and a set of Agent jobs: DatabaseBackup, DatabaseIntegrityCheck and IndexOptimize. It backs up every current and future database, names files by server, database, type and timestamp, cleans up by age, supports CHECKSUM, compression, striping and backup to URL, and only rebuilds indexes that need it. It is what most DBAs install on day one. A full backup of all user databases, kept for 14 days, is:

EXECUTE dbo.DatabaseBackup
  @Databases = 'USER_DATABASES',
  @Directory = 'D:\SQLBackup',
  @BackupType = 'FULL',
  @Compress = 'Y',
  @CheckSum = 'Y',
  @Verify = 'Y',
  @CleanupTime = 336;

Install the script, then set the schedules on the jobs it creates (they are created without schedules). Whichever route you choose, make the jobs email or log a failure somewhere a human looks: a backup job that has silently failed since March is the most common finding in a post-incident review.

Off the server: BACKUP TO URL

Backups on the same server as the database share its fate. SQL Server can write directly to object storage with BACKUP TO URL: Azure Blob Storage since 2012, and since SQL Server 2022 any S3-compatible endpoint, which includes the S3-compatible object storage offered by hosting providers and by MinIO. The credential is created once, then every backup can target the bucket:

-- S3-compatible storage (SQL Server 2022 and later)
CREATE CREDENTIAL [s3://s3.example.net:9000/sqlbackups]
WITH IDENTITY = 'S3 Access Key', SECRET = 'ACCESSKEY:SECRETKEY';

BACKUP DATABASE shop
TO URL = 's3://s3.example.net:9000/sqlbackups/shop_FULL_20261106.bak'
WITH COMPRESSION, CHECKSUM, FORMAT, MAXTRANSFERSIZE = 20971520;

S3 backups are limited to 10,000 parts per file, so large databases are striped across several URLs (up to 64) and MAXTRANSFERSIZE is raised to 20 MB; the Microsoft Learn page on S3-compatible object storage has the arithmetic, as read on 6 November 2026. For Azure Blob Storage the credential uses a shared access signature and the URL starts with https. Ola Hallengren's DatabaseBackup supports both with the @URL and @Credential parameters.

If BACKUP TO URL is not an option (Express editions before 2022, or a provider without S3), the alternative is a scheduled copy: back up to local disk, then a PowerShell step with Copy-Item to an SMB share, rclone or the AWS CLI to object storage, running after the backup job. Encrypt backups that leave the server, either with WITH ENCRYPTION in the backup command (requires a server certificate, which you then also have to back up and keep) or at the storage layer. On MassiveGRID, backup storage is available alongside the servers, and the managed SQL Server host's daily VM backups cover the operating system and the local backup folder as well; your off-server copy is still the one that satisfies an auditor.

Retention, and what to delete when

Keep backups long enough to cover the slowest way you discover a problem. A deleted row is found in a day; a corrupt page or a subtle data bug can take weeks. A common scheme for a single application:

Delete by age with the cleanup task or @CleanupTime, and never delete a full backup while a differential or log chain still depends on it. Document the scheme in one paragraph and store it with the restore runbook; the backup retention policy guide goes deeper on the trade-offs. GDPR and similar regimes do not forbid keeping personal data in backups, but they expect a stated retention and a way to honour erasure when the backup is restored, so write that down too.

The monthly restore drill

A backup that has never been restored is a hypothesis. Once a month, restore the latest full, differential and a few log backups to a second instance, run DBCC CHECKDB on the result, time the whole thing, and write the time down: that number is your recovery time objective, measured rather than guessed. The restore itself, to a copy with a new name and relocated files:

RESTORE DATABASE shop_drill FROM DISK = 'D:\SQLBackup\shop_FULL.bak'
WITH MOVE 'shop' TO 'D:\SQLData\shop_drill.mdf',
     MOVE 'shop_log' TO 'D:\SQLLog\shop_drill.ldf',
     NORECOVERY, CHECKSUM, STATS = 10;
RESTORE DATABASE shop_drill FROM DISK = 'D:\SQLBackup\shop_DIFF.bak' WITH NORECOVERY, CHECKSUM;
RESTORE LOG shop_drill FROM DISK = 'D:\SQLBackup\shop_LOG_1400.trn' WITH NORECOVERY;
RESTORE LOG shop_drill FROM DISK = 'D:\SQLBackup\shop_LOG_1415.trn' WITH STOPAT = '2026-11-06 14:06:00', RECOVERY;
DBCC CHECKDB ('shop_drill') WITH NO_INFOMSGS;
DROP DATABASE shop_drill;

A cheap second instance for drills is a Windows VPS with SQL Server Express if the database fits under 10 GB, or the 180-day evaluation edition on a small VPS for larger ones (the evaluation is not for production, but a monthly drill server that is rebuilt from the backup is exactly what it is for). On the managed host, ask MassiveGRID for a clone of the server from the daily VM backup and run the database restore on the clone; the VM backup gets you a machine with the right SQL Server build in minutes, and the database backup then proves itself on it.

While you are there, test the two scenarios the schedule exists for: restore to a point in time just before a deliberate DELETE, and restore a single table by pulling it from the drill copy with an INSERT ... SELECT. If either takes longer than the business can stand, the fix is in the schedule, not in the day of the incident.

Frequently Asked Questions

If my hosting provider backs up the VM every day, do I still need SQL Server backups?

Yes. The VM backup restores the whole server to one point a day and protects against a lost disk or a broken operating system. Only SQL Server's own full, differential and log backups give you point-in-time recovery, restore to another server, a per-table rescue, and a checksum pass over every page. Run both, and keep a copy of the database backups off the server.

How often should I take transaction log backups?

Every 15 minutes is the usual starting point for a production database in the Full recovery model, and every 5 minutes where the data is valuable enough. On a quiet system each log backup is tiny, so a frequent schedule costs almost nothing; what it buys is the recovery point objective. If you will not take log backups at all, switch the database to the Simple model so the log does not fill the disk.

Can SQL Server Express take scheduled backups?

Yes, but Express has no SQL Server Agent, so schedule a sqlcmd or PowerShell script with Windows Task Scheduler. Full backups with compression and checksum work in Express, and BACKUP TO URL to S3-compatible storage is available from SQL Server 2022 Express. Ola Hallengren's procedures run on Express as well when called from Task Scheduler.

Should I use WITH COMPRESSION and WITH CHECKSUM on every backup?

Yes. Compression typically halves backup size and time on normal data at a small CPU cost, and it is available in every edition from SQL Server 2016 SP1. CHECKSUM makes the backup verify each page as it is read and adds a backup checksum you can verify with RESTORE VERIFYONLY, so a corrupt page is caught at backup time rather than at restore time.

How are backups handled on MassiveGRID SQL Server hosting?

The Managed Cloud Dedicated Server behind the SQL Server hosting page is backed up daily as a VM with on-demand snapshots, and its storage is replicated three times on the HA cluster. Database-level backups, their schedule, retention and an off-server copy remain yours to configure, because only you know each database's recovery point objective; this guide is the setup MassiveGRID recommends.

Daily VM backups, replicated storage, and room for your own SQL backups

MassiveGRID SQL Server hosting runs on a Managed Cloud Dedicated Server with Windows Server Standard and SQL Server Web licensed per vCore, from $86.34 a month for 2 reserved vCores, 8 GB and 128 GB NVMe. Daily VM backups with snapshots, 3x replicated storage and HA failover are included; add NVMe or backup storage for your database backup chain in New York, London, Frankfurt or Singapore.

See SQL Server hosting

Further Reading