Servers

Storage layouts

How identical disks are combined in RAID: No RAID, RAID 0, 1, 5 and 10. What each gives in space and safety, and what happens when a disk dies.

Disks wear out and fail. A storage layout (RAID) decides how a server's identical disks work together, and therefore what one dead disk does to your customers.

The layouts

  • No RAID (JBOD): every disk on its own. All the space, but a dead disk takes its data with it.
  • RAID 0 (stripe): the disks act as one big, fast disk. All the space, but if any disk dies, everything is lost. It makes failures worse, not better.
  • RAID 1 (mirror): the disks hold exact copies of each other. You get the space of one disk, and all but one can die without losing data.
  • RAID 5: needs at least three disks. You give up one disk's space; any one disk can die safely.
  • RAID 10: needs at least four disks, an even number. Half the space, fast, and survives a disk failure in each mirrored pair.

The panel shows, for your disks, the usable space of each layout and how many failures it tolerates. Layouts your disks cannot form say needs more disks.

Identical disks

Only identical disks (same type, size and form) form a layout. Mixed disks each stay on their own. The one exception is an array you are growing onto larger disks, below.

Growing an array with larger disks

A mirrored or parity array (RAID 1, 5 or 10) can get bigger while the server keeps running, the way it is done on a real server: you swap its disks one at a time for larger ones of the same kind.

  1. Buy a larger disk of the same kind (a 1.92 TB SAS SSD for an array of 960 GB SAS SSDs, for example).
  2. On the server's page, under Storage layout, click Swap for a larger disk, pick the disk to pull and the larger spare, and confirm.
  3. The array rebuilds onto the new disk for a few minutes. Customers stay online and the RAID level stays the same.
  4. Repeat for every other disk in the array.

Until the last disk is swapped, the array runs at its smallest disk's size, so the extra space only arrives with the final swap. The disks you pull go back to your inventory: sell them if you bought them new, or keep one as a spare.

While a rebuild runs the array has one copy fewer. If another disk dies during it, a mirror can lose its data, so do it with every other disk healthy, and keep a backup.

A plain disk without RAID cannot be swapped like this: its data has no second copy. Move the customers off the server first (moving virtual machines), or add a disk to a free bay.

When a server with a mirrored array is running short of disk space, its Resource use panel reminds you of this. If the disks are out of speed rather than space, larger disks of the same kind do not help: move some virtual machines to another server or use faster disks.

When to choose

Choose the layout before installing the operating system. Changing it later rebuilds the disks and wipes the server, operating system included.

When a disk dies

  • Inside what the layout tolerates: the server keeps running, marked degraded. Replace the disk soon; disks are hot-swap, so the server stays up.
  • Beyond it: the array is lost. Virtual machines are destroyed, the operating system is wiped and customers lose their data, unless the server has a backup.

Tip: For servers with customers, a layout that survives a dead disk plus a spare disk in stock is the cheapest insurance you can buy.

Common mistakes

  • RAID 0 for customer data.
  • Changing the layout on a live server and wiping it, when swapping the disks one at a time for larger ones would have kept every customer online.
  • Buying a larger disk for a server whose bays are all full, and expecting it to fit: swap it in for one of the old disks instead.
  • Waiting days to replace a disk in a degraded array.

Want the numbers? The full rules: Building servers

Start a hosting company with $50,000

Free to play in your browser. One shared world, running in real time.