Block storage is a way of providing storage in which capacity is presented to a server as raw volumes made up of fixed-size blocks, each with an address but no meaning of its own. The server’s operating system sees the volume as a disk, formats it with a file system or hands it to a database, and decides how data is organized. Block storage is the model behind local hard drives, storage area networks (SANs) and the disks attached to cloud virtual machines, and it is the usual choice for databases and other performance-sensitive applications.
At a glance
- Storage is delivered as volumes, also called disks or LUNs, that a server uses as if they were its own drives.
- The server, not the storage system, manages files and folders on the volume.
- Typically offers low latency and high IOPS, which suits databases and virtual machine disks.
- In the cloud, sold per gigabyte provisioned, in performance tiers, often with snapshots billed separately.
- One of three main storage models, alongside file storage and object storage.
What problem it solves
Applications such as databases, ERP systems and virtual machines read and write small pieces of data constantly and need each operation to finish quickly and predictably. They also expect to control their own disk layout. Storage that adds a shared file layer or a web API in between can introduce latency and overhead these workloads don’t tolerate well.
Block storage gives them direct, disk-like access. Because the storage system only deals with blocks, it can be very fast, and it can be carved into volumes sized to each need, moved between servers, snapshotted and replicated. Separating block volumes from the server also means a failed server can be replaced and the same data volume attached to the new one, which is a basic building block of virtualization and cloud computing.
How it works
Volumes and blocks. The storage system pools its physical drives and divides capacity into volumes. Each volume is a sequence of blocks with addresses. The server reads and writes blocks by address, and its file system or database keeps track of what each block holds.
Connecting servers to storage. Block storage can be local, with disks inside the server; networked, through a SAN using protocols such as Fibre Channel, iSCSI or NVMe over Fabrics; or virtualized, where a hypervisor or cloud platform presents a virtual disk backed by a distributed storage system. In public cloud, you create a volume, choose its size and performance tier and attach it to an instance.
Performance tiers. Flash-based tiers deliver higher IOPS and lower latency; disk-based tiers cost less for large sequential workloads. Many cloud services let you provision IOPS and throughput separately from capacity, or scale performance with volume size.
Data protection. Storage systems usually protect against drive failure with RAID, erasure coding or replication. Snapshots capture point-in-time copies of a volume, and many systems can replicate volumes to another site. Snapshots alone are not a full backup unless they are copied somewhere independent of the original storage.
See our file and object storage and public cloud pages for how buyers compare storage options across environments.
When it matters for buyers
- Sizing cloud virtual machines. Disk type and size often affect application performance and monthly cost as much as the instance does.
- Database performance problems. Storage latency and IOPS limits are a common cause; check the volume tier before upgrading compute.
- Cost reviews. Overprovisioned volumes, detached volumes and old snapshots are frequent sources of cloud waste.
- Storage refreshes. On premises, block storage comes from SAN arrays, hyperconverged clusters or storage as a service offerings with different cost and management trade-offs.
- Choosing managed services. Database as a service offerings still sit on block storage, and their storage performance limits are worth checking.
Questions to ask vendors
- What IOPS, throughput and latency does each tier deliver, and are those figures sustained or burst?
- Is pricing based on provisioned capacity, used capacity or both, and how is performance priced?
- How are snapshots stored and billed, and can they be copied to another region or account?
- Can volumes be resized or changed to another tier without downtime?
- What redundancy protects a volume against hardware failure, and what is the durability commitment?
- Is data encrypted at rest, and who controls the keys?
How it differs from file and object storage
File and object storage are the other two main models. File storage, such as a NAS, manages its own file system and shares named files and folders with many users and servers at once. Object storage keeps data as objects with metadata in a flat pool reached through a web API, and scales to huge capacities at low cost. Block storage does neither: it hands a server raw disk space and leaves organization to the server. That makes it typically the fastest for databases and virtual machine disks, but harder to share and more expensive per gigabyte than object storage. Most environments use all three for different jobs.