Adding a Second Data Disk to a Windows VPS

TL;DR / Summary: A Windows VPS usually ships with one system disk, and everything ends up on it. Attaching a second volume, initializing it as GPT, and moving application data onto it takes about twenty minutes and makes snapshots, rebuilds, and cleanup far less frightening.

When the system disk should not hold data

Keeping everything on C: works until you need to rebuild the server. Then the data has to be copied out through RDP or a temporary network share, usually under time pressure and often badly.

A separate data volume changes the failure model. The system disk can be replaced, reimaged, or rolled back while the data disk stays attached and untouched.

It also makes snapshots honest. You can snapshot the operating system without dragging hundreds of gigabytes of customer files into every image.

Attaching and initializing the disk

Attach the volume in your provider’s control panel first. Some platforms require a reboot or a rescan before Windows notices it; a quick way to force the rescan is Get-Disk in an elevated PowerShell session.

Open Disk Management (diskmgmt.msc) and look for the new disk marked as offline or unallocated. Right-click to bring it online, then initialize it as GPT unless you have a specific reason to choose MBR.

GPT is the sensible default on any modern Windows Server image. It handles large volumes, avoids the four-partition limit, and is what the recovery tools expect.

Drive letters, folders, and mount points

A drive letter is the easiest choice, and D: is the conventional pick because the system disk has already taken C:. Avoid letters that mapped network drives might want later.

A folder mount point is tidier for application data: create an empty folder such as C:\Data and mount the volume there. The path stays stable even if the drive letter changes, which matters on servers where other software grabs letters without asking.

Whichever you choose, document it. The next person to log in will not guess that the data lives on a mounted folder rather than a drive.

Moving application data without breaking paths

Stop the service before you copy. A database or a mail store that is actively writing will produce a corrupt copy, and the corruption may not show up until the next restart.

Use robocopy with /MIR for the first pass, then run it again with /MIR after the service is stopped to catch the changes. Two passes are faster than one very slow pass and much safer than copying a live directory.

After the copy, repoint the application at the new path. Where software hardcodes a location, a directory junction can bridge the gap, but treat junctions as a temporary measure and note them in your documentation.

Run the second copy pass with the service still stopped, compare the file count at both ends, and only then start the service. A count that differs by a handful of files usually means an antivirus scanner or a log writer touched the old directory, and it is worth resolving before anyone declares the move finished.

Formatting choices for a data volume

NTFS remains the right answer for almost every workload. ReFS is worth considering for large file stores that benefit from integrity streams, but plenty of applications still refuse to install on it.

Use 64K allocation units for volumes that hold large media files or database backups, and the default 4K for volumes full of small documents. The wrong size wastes space; the right one is rarely dramatic.

A quick format is fine for a brand new disk. A full format on a multi-terabyte volume takes hours and only matters when you are checking for bad sectors on used hardware.

Confirming the disk survives a reboot

Reboot before you move anything important. A disk that appears online but is not set to come online automatically can be missed by services that start at boot, and the resulting error looks like a database failure rather than a storage one.

Check the drive letter or mount point after the reboot, then check that the service started and can read its files. Do this while the server is still idle and you still have time to fix it.

Add the volume to your monitoring. Free space on C: is usually watched; free space on the data disk is often forgotten until it fills.

Resizing and expanding the data volume later

Most providers let you grow a volume but not shrink it. Extending is straightforward in Disk Management once the provider has increased the size and Windows has rescanned the disk.

Leave a little unallocated space at the end of the volume if you expect to grow again soon, and never extend a volume that holds a live database without a backup first. A resize interrupted by a full disk is one of the few storage operations that can lose data outright.

Two disks, one clean separation

The system disk is disposable; the data disk is the part you protect. Once that separation exists, rebuilding a Windows VPS becomes a routine task instead of an evacuation.

Attach the volume, initialize it, decide between a letter and a mount point, copy with the service stopped, and reboot to prove it holds. Four steps, one afternoon, and a much calmer server.

If you have not chosen a plan yet, the Windows VPS hosting guide covers how much disk a small workload actually needs, which makes the second volume easier to size.

Start with the smallest data disk that covers a year of growth and grow it later. Storage is the easiest part of a VPS to expand, provided you set it up separately in the first place.