Data Protection Montana: How to Build a Compliance-Ready On-Premises Backup and Retention System

Compliance-ready on-premises backup system with Montana mountain silhouettes and state outline

A small Montana business can build a defensible backup and retention system without paying recurring per-user cloud fees. This guide explains how to inventory regulated data, create retention rules, deploy an owned Linux backup host, protect repositories from ransomware, control access, and document restore tests.

The goal is not to claim that hardware alone creates compliance. Compliance depends on documented decisions, appropriate safeguards, repeatable procedures, and evidence that those procedures operate as written.

Important: Retention requirements vary by industry, contract, state law, and the type of record involved. Use this guide as a technical planning framework and have legal or compliance professionals review the final policy.

Why retention and audit trails are where small businesses fail

Many businesses can show that a backup job runs. Fewer can answer these questions:

  • What data is included in the backup?
  • How long is each record type retained, and why?
  • Who can delete a backup?
  • Can the business identify who restored a file?
  • When was the last full restore test?
  • What happens when a customer requests deletion?
  • Where are the logs, approvals, and test results stored?

That evidence is what auditors, insurers, and business partners typically want to see. Useful evidence includes a dated retention policy, data inventory, backup configuration, job completion logs, access logs, deletion approvals, restore-test reports, encryption-key custody records, and a list of storage media.

The Montana Consumer Data Privacy Act does not establish one universal retention period for every personal record. Instead, it emphasizes data minimization, purpose limitation, reasonable safeguards, and retaining information only as long as reasonably necessary for the disclosed purpose. HIPAA and PCI DSS introduce additional expectations for certain organizations. HIPAA documentation generally requires six years of retention, while PCI DSS requires at least one year of audit-log history, with three months immediately available for analysis.

Those rules do not mean every file must be kept for six years. They mean the business must classify information and document the reason for each retention period.

Start with a data inventory and retention schedule

Before buying disks, list the systems and record types that require protection:

  • Customer and employee personal information
  • Accounting, tax, payroll, and contract records
  • Protected Health Information (PHI), if applicable
  • Payment-card data and systems connected to the cardholder data environment
  • Email, file shares, databases, and line-of-business applications
  • Security, operating-system, database, and administrator logs
  • Records of deletion requests, incidents, and legal holds

Then assign an owner and retention rule to each category. A practical starting schedule might look like this:

Data or record type Example retention approach Evidence to retain
Daily operational data 35–90 days Backup reports and snapshot listings
Weekly business backups 12–13 months Job logs and repository inventory
Tax, contract, and financial records Often seven years, subject to review Policy justification and deletion approval
HIPAA-related documentation At least six years Policy versions, training, risk and incident records
Security logs One year for PCI-related activity; longer where HIPAA or insurance requirements apply Central log reports and access history
Cardholder data As short as operationally necessary; tokenize or avoid storing it Data-flow diagram and disposal records

These are planning examples, not automatic legal requirements. The policy should explain the purpose behind each period and identify the person authorized to approve changes or deletion.

Montana business using open-source infrastructure to reduce IT costs

Build the owned on-premises backup stack

A small organization can use a server it owns rather than a per-seat backup platform. The one-time capital cost includes the host, memory, disks, an uninterruptible power supply, and replacement media. Ongoing costs still exist for electricity, hardware replacement, maintenance, and offsite storage, but the design avoids recurring licensing tied to every employee.

A typical stack contains:

  1. Linux host: Proxmox for virtual machines or Debian for a dedicated backup server.
  2. Redundant storage pool: ZFS or Btrfs with checksums and snapshots.
  3. Encrypted backup tool: Restic or Borg.
  4. Offsite destination: A second owned server, removable disk set, or another controlled location.
  5. Central logs: Local syslog or Grafana Loki.
  6. Key custody: Documented encryption keys stored separately from the backup host.

For a Debian host, install the required tools and create a dedicated backup service account:

sudo apt update
sudo apt install zfsutils-linux restic openssh-server rsyslog
sudo adduser --system --group --home /var/lib/backup backupsvc
sudo install -d -o backupsvc -g backupsvc /srv/backup-repo

ZFS is useful when snapshot management, checksumming, and replication are priorities:

sudo zpool create datapool mirror /dev/disk/by-id/DISK_A /dev/disk/by-id/DISK_B
sudo zfs create datapool/backups
sudo zfs set compression=lz4 datapool/backups
sudo zfs set atime=off datapool/backups

Do not copy these device names without confirming them. Selecting the wrong disk can destroy existing data.

Btrfs is another option when the administrator prefers a Linux-native filesystem with subvolumes and snapshots:

sudo mkfs.btrfs -L backup-pool /dev/sdb
sudo mount /dev/sdb /srv/backup-repo
sudo btrfs subvolume create /srv/backup-repo/repositories

Use enterprise-grade disks, monitor their health, and leave free capacity for snapshot growth. RAID or mirroring protects against some disk failures; it does not protect against ransomware, fire, theft, or accidental deletion.

Diagram suggestion: show production servers feeding encrypted Restic or Borg backups into a local ZFS pool, then replicating to a physically separate offsite repository.

Add encrypted, versioned backups

Restic and Borg both encrypt backup data before storage. They also support deduplication, which reduces storage consumption when many files share unchanged blocks.

A simple Restic repository initialization looks like this:

sudo -u backupsvc restic -r /srv/backup-repo init

A backup job can then capture selected production paths:

sudo -u backupsvc restic -r /srv/backup-repo backup \
  /srv/business-data /etc /var/lib/postgresql

Use a password file with permissions restricted to the backup account rather than placing the repository password directly in scripts:

sudo install -o backupsvc -g backupsvc -m 600 /dev/null /etc/restic-repo.pass

The password file is not a complete key-management plan. Keep a second protected copy of the recovery key in a separate location, document who can access it, and test that it works. If the only copy exists on the backup server, a server failure can make the backups unusable.

For offsite storage, use a second repository or a replicated ZFS dataset. With Restic, the backup set can be copied to another repository:

restic -r /srv/backup-repo \
  --repo2 sftp:backup-offsite:/srv/restic \
  copy

The offsite system should not be continuously writable by every production administrator. A separate network, account, firewall rule, or scheduled connection reduces the chance that ransomware reaches both copies.

Use append-only repositories and controlled pruning

An append-only repository allows backup data to be added while preventing ordinary backup clients from deleting or modifying existing snapshots. For example, a Borg service can be restricted with an append-only command:

command="borg serve --restrict-to-repository /srv/borg/repo --append-only",no-port-forwarding,no-agent-forwarding,no-pty ssh-ed25519 AAAA...

Restic can also use an append-only server mode. The important operational distinction is that normal backup accounts cannot prune. A separate maintenance account performs approved retention cleanup during a scheduled maintenance window.

Example retention logic:

restic forget \
  --keep-daily 35 \
  --keep-weekly 13 \
  --keep-monthly 84 \
  --prune

Do not run a command like this until the policy, repository backups, and approval process are in place. A mistaken retention rule can permanently remove records that the business is required to keep.

For immutable ZFS snapshots, use a controlled naming convention and an approved destruction script. Store the script in version control, require a second person to approve long-term deletion, and log the request, operator, timestamp, record category, and resulting snapshot names.

Secure infrastructure represented as a fortified digital environment with Montana mountains

Separate access, keys, and audit logs

Do not let employees use a shared administrator password for backup operations. Create separate accounts for:

  • Backup writing
  • Repository maintenance and pruning
  • Restore operations
  • System administration
  • Log review

Use SSH keys rather than password-based automated jobs. Restrict each key by source address and command where possible. A backup-writing key should not be able to open an interactive shell or delete snapshots.

Ship authentication, backup, restore, and deletion events to a protected logging host. Local syslog is sufficient for a small deployment:

# /etc/rsyslog.d/30-backup.conf
if $programname == 'restic' then /var/log/backup/restic.log
& stop

Grafana Loki is another self-hosted option for searchable log retention. Whichever tool is selected, configure alerts for failed jobs, repository checks, unexpected pruning, and restore activity.

The logs should answer: who performed the operation, what repository or dataset was affected, when it happened, whether it succeeded, and which ticket or approval authorized it.

Document deletion, legal holds, and evidence storage

The written policy should identify:

  • Systems and data categories in scope
  • Business or legal purpose for each category
  • Minimum and maximum retention periods
  • Snapshot and backup schedules
  • Who approves deletion
  • How legal holds suspend deletion
  • How customer deletion requests are handled
  • Where keys, logs, policies, and test reports are stored
  • How exceptions are documented

When a deletion request arrives, delete or anonymize the record in the production system first. Keep only the minimum information needed to document the request and prevent the data from being reintroduced. Do not manually edit immutable backups. Instead, document that backup copies age out under a defined, time-limited policy.

If litigation, an investigation, or an insurance claim is active, place the affected records under a legal hold and suspend normal deletion until the hold is released by the authorized person.

Test restores and record the result

A backup that has never been restored is an assumption, not evidence. Test at least one file restore monthly and a complete application or server recovery quarterly. For critical systems, perform a recovery into an isolated network so the test does not overwrite production data.

A restore report should include:

  • Date and operator
  • Backup snapshot or archive selected
  • Files, database, or virtual machine restored
  • Recovery time and result
  • Hash or application validation where practical
  • Problems found
  • Corrective action and completion date

For example:

restic -r /srv/backup-repo snapshots
restic -r /srv/backup-repo restore latest --target /srv/restore-test
sha256sum /srv/restore-test/critical-file.pdf

Keep the report with the backup policy and job logs. Auditors and cyber-liability insurers often ask for restore evidence before they ask for a diagram.

Montana cybersecurity operations environment with monitoring, servers, and security controls

Understand the limits of an on-premises design

A single server in one Montana office is not a complete disaster-recovery plan. Flooding, fire, theft, extended power loss, or a regional event can destroy the production system and its local backup. Maintain at least one offsite copy, verify it regularly, and rotate removable media if that is part of the design.

Other risks include silent disk failure, corrupted snapshots, forgotten encryption keys, unsupported Linux packages, inadequate cooling, and retention rules that grow storage faster than expected. Estimate capacity using actual change rates, not just current data size.

Outside help is justified when the business handles sensitive health data at scale, stores payment-card data directly, operates multiple locations, lacks staff who can maintain Linux and storage systems, or cannot perform independent restore tests. The cost of a poorly implemented system includes lost records, failed insurance reviews, extended downtime, and an inability to prove what happened.

For related background on owned infrastructure, see The Cloud vs. In-House IT Infrastructure: A Pragmatic Approach. The technical design should always follow the written retention policy, and the policy should be reviewed whenever the business changes its systems, data, contracts, or regulatory obligations.

Call us at 406-866-0128 or visit https://itgreatfalls.com.