Storage accounts, storage services, tiers and redundancy

The account, and the services inside it

A storage account is the container for the settings and the namespace: its name is globally unique and forms the hostname other things address it at, and the redundancy option, the default access tier, the network rules and the encryption settings are all properties of the account rather than of the data inside it. Managed disks are the exception worth knowing — they are their own resource type and do not sit in an account you create.

Four services live inside, and the quickest way to pick between them in a scenario is to ask how the application wants to ask. Blobs are objects fetched over HTTP by URL, which suits unstructured data at any scale but means an application expecting a file path would have to be rewritten. Azure Files provides managed shares mounted over SMB — and over NFS on the premium tier — so an existing application carries on using paths, and on-premises servers can mount the same share. Queues carry messages so components can work independently of each other, and tables hold schemaless key-value rows. A path or a URL is usually the whole discriminator.

Redundancy: count the copies, then ask where they are

Locally redundant storage keeps three copies inside a single datacentre, which protects against a failed disk or rack and nothing larger. Zone-redundant storage writes three copies across availability zones in the primary region, so losing a whole datacentre is transparent to the application. Geo-redundant storage adds a second location entirely: three copies locally plus three in the paired region. Geo-zone-redundant combines the two, spreading the primary copies across zones and still keeping a copy in the pair.

The synchronous-against-asynchronous distinction matters as much as the geography. In-region replication completes before the write is acknowledged, so those copies are never behind. Cross-region replication is continuous but asynchronous, so a regional failover can lose the most recent writes — which is what a recovery point objective is describing. And redundancy is not readability: with plain geo-redundant storage the second copy exists for recovery and cannot be read, and reading it needs either the read-access variant, which gives you a second read-only endpoint, or a failover.

Access tiers trade storage price for access price

Pick a tier with two questions: how often is this read, and how long can you wait for it. Hot has the highest storage price and the lowest access price, which suits data read constantly. Cool and cold are progressively cheaper to keep and dearer to read, and both are still online, so data comes back in milliseconds. Archive is offline: the storage price is the lowest available and a blob has to be rehydrated to an online tier before it can be read, which takes hours rather than moments.

Two details that decide questions. Archive is set on individual blobs rather than as an account default, unlike the online tiers. And each of the cooler tiers carries a minimum retention period — thirty days for cool, ninety for cold, a hundred and eighty for archive — so deleting or moving data early triggers an early-deletion charge that can undo the saving you moved it for. Where consistently low latency and high transaction rates are what matter, a premium block blob account is the answer and it costs accordingly.