CloudCostCompareBlock storage volume types and what each really costs
Block storage is the disk a server boots from and keeps its data on. It is separate from the server: you can detach a volume and attach it elsewhere, and it keeps billing while it exists, whether or not a server is running.
What the names are
| Cloud | Product |
|---|---|
| AWS | Elastic Block Store (EBS) |
| Azure | Managed Disks |
| Google Cloud | Persistent Disk and Hyperdisk |
| Oracle Cloud | Block Volume |
You pay for size, not use
A volume is billed for the capacity you provision, per gigabyte per month, even if it is mostly empty. A 500 GB disk holding 20 GB costs the same as one holding 480 GB. This is the opposite of object storage, where you pay for what you store. See object storage classes and lifecycle.
Performance tiers
Volumes come in tiers that trade price for speed. The names vary, but the pattern is steady.
- Magnetic or HDD tiers: low price per gigabyte, suited to large sequential data such as logs and archives, not to databases.
- General purpose SSD: the default for most servers and databases.
- Provisioned performance SSD: you pay for a guaranteed rate of operations per second, for demanding databases.
Within a tier, a vendor may charge separately for performance above a baseline. On AWS, a general purpose gp3 volume includes a baseline of operations and throughput, and you pay for provisioned amounts above it. Some vendors tie performance to size, so a larger disk is also a faster disk.
What is not in the per-gigabyte price
- Provisioned IOPS and throughput above the included amount.
- Snapshots, which are billed by the amount stored. See backup and retention costs.
- Cross-region copies, which add transfer charges.
- Encryption is usually free, though a managed key service may charge for key use.
Our storage catalog lists capacity prices per gigabyte for each volume type we have read from the vendor’s price list, and the page says which charges it leaves out. Treat the table as the floor, not the whole cost.
Choosing a type
- Start with the default general purpose SSD. It is right for most things.
- Look at the real demand. Operations per second and throughput are metrics you can read for a week.
- Pay for provisioned performance only when the metrics show you need it.
- Use a cheaper tier for data that is large and rarely hot, such as old logs.
- Size generously but not wildly. Growing a volume is usually easy, while shrinking one usually is not.
Common waste
Disks left behind when a server is deleted, snapshots of those disks, and volumes sized for a data growth that never came. A monthly review of unattached volumes is among the cheapest savings available. The wider routine is in the cost optimization checklist.
For databases specifically, see managed database versus self-hosted, where storage is one line of several.