Skip to content

Independent marketplace for verified cloud accounts

Block, Object and File Storage Explained

Three storage models with different access patterns, durability properties and billing. This guide explains how each works and which one fits common workloads.

Cloud storage comes in three shapes and they are not interchangeable. Choosing the wrong one does not usually produce an error message. It produces a system that works in testing, costs more than expected, and becomes awkward to change once there is real data in it.

The three are block, object and file. They differ in how data is addressed, what guarantees they offer, how they are billed, and what they are physically good at. This guide covers each in turn, then puts them side by side on the dimensions that matter when you are deciding.

Why three types exist at all

They exist because applications ask for data in fundamentally different ways. A database wants to read and write small fixed size chunks at arbitrary offsets, very quickly, with no other writer involved. A media application wants to fetch a whole file over the network and does not care about offsets. A legacy application wants a directory tree that several machines can mount at once.

No single storage design serves all three well. Optimising for random access at low latency means attaching storage to one machine over a fast path. Optimising for cheap, massive, durable capacity means spreading data across many machines and accepting that you address it as whole objects. The three types are what you get when you follow each requirement to its conclusion.

Block storage, a disk attached to a machine

Block storage presents as a raw disk. You attach a volume to an instance, format it with a filesystem, and the operating system treats it like local hardware. The storage layer knows nothing about files; it serves fixed size blocks and leaves interpretation to the filesystem above.

That is what makes it fast and what makes it restrictive. Typically a volume attaches to one instance at a time, so it does not solve sharing. It has a fixed provisioned size, which you pay for whether or not it is full, and growing it is an operation rather than an automatic behaviour. Snapshots are the usual backup mechanism and are generally incremental after the first one.

Use block storage for anything that needs a filesystem with real write performance: databases, message brokers, build workspaces, and operating system root volumes. Performance is usually sold as a tier or as provisioned throughput, and the specifications are easy to misread, which is why reading storage performance specs is worth a separate look.

Object storage, buckets and keys over HTTP

Object storage drops the filesystem entirely. You store an object under a key in a bucket, and you read and write it over an HTTP API. There are no directories, only key names that often contain slashes and look like paths. You generally cannot modify part of an object in place; you replace the whole thing.

In exchange you get capacity that is effectively unbounded, data replicated across multiple facilities by default, and billing that charges for what you actually store rather than what you provisioned. Objects can be served directly to clients, versioned, lifecycle managed into colder tiers, and given expiry rules.

The costs that surprise people are not the storage ones. Requests are billed per operation, which makes many tiny objects disproportionately expensive. Retrieval from archival tiers carries its own charge and its own delay. And data leaving the provider is billed separately on most platforms, which is the mechanism behind a great many unexpected invoices. Use object storage for backups, media, logs, static assets, data lakes and anything written once and read many times.

File storage, a shared mount

File storage gives you a filesystem over the network, mounted by several machines at once using a standard protocol. It preserves directories, permissions and file locking semantics, which is exactly what applications written before object storage existed expect to find.

Its value is compatibility and sharing. If three application servers need to see the same uploads directory, file storage solves it with no code change. The costs are latency, since every operation crosses the network, and price, since managed file services are usually the most expensive of the three per unit of capacity.

Use it for shared content between instances, home directories, and applications that genuinely need POSIX semantics. If you are building something new and can write to an object API instead, that is almost always the cheaper long term choice.

Durability, availability and redundancy

These three words get used loosely and they mean different things. Durability is the probability that stored data survives. Availability is the probability that you can reach it right now. Redundancy is the mechanism, such as replication across facilities, that produces the other two.

Object storage is generally the most durable of the three because replication across facilities is the default rather than an option. Block volumes are typically replicated within one facility, which protects against a disk failure but not against the loss of that facility. File services vary by product and by the tier you select.

None of this is backup. Replication faithfully copies a deletion or a corruption to every replica. Backup requires a separate copy with its own retention, ideally in a different account so that a compromised credential cannot destroy both, which is one of the practical arguments in the case for account isolation. Design your recovery around the failure you actually fear, then test the restore rather than assuming it.

What each type costs you

The billing models differ enough that comparing headline capacity prices is meaningless.

Type Billed on Easy to overspend on
Block Provisioned capacity, plus performance in some tiers Large volumes that are mostly empty, and forgotten snapshots
Object Stored bytes, requests, and retrieval from colder tiers High request counts and premature use of archival tiers
File Capacity, sometimes with a throughput component Using it as a general dumping ground because it is convenient

Data transfer sits on top of all three and is charged separately by most providers. It is the single most common reason a storage bill exceeds the estimate, and the reason is usually architectural rather than wasteful. Egress fees explained sets out where the charges come from. Providers also differ sharply here: several European and value oriented hosts bundle a transfer allowance rather than metering every gigabyte, which shows up clearly in a comparison such as Hetzner against OVHcloud.

Performance characteristics worth checking

For block storage, the numbers that matter are operations per second, throughput in megabytes per second, and latency. Which one limits you depends on your access pattern: a transactional database is usually bound by operations and latency, while a sequential job is bound by throughput. Many providers tie performance to volume size, so a small volume can be slow for reasons unrelated to the storage tier you picked.

For object storage, per request latency is higher and fairly constant, and you scale by parallelism rather than by provisioning. For file storage, expect network latency on every metadata operation, which is why workloads that touch many small files feel slow even when raw throughput looks fine. Test with your real access pattern rather than a generic benchmark, because the gap between the two is routinely large.

Picking a type for common workloads

A relational or transactional database belongs on block storage, with snapshots and a tested restore. Static assets and media belong in object storage, served through a cache. Application logs belong in object storage with a lifecycle rule, not on the instance disk where they will eventually fill the root volume.

Backups belong in object storage, in a separate account where possible. Shared uploads between several application servers go to object storage for new systems and file storage for older ones that expect a mount. Machine learning training data usually lives in object storage and is staged onto block storage near the accelerators, a pattern covered in data pipelines for machine learning.

Moving data between types later

Assume you will get at least one of these decisions wrong, and plan for the correction. Moving from block to object is the common direction and it means changing application code, since a filesystem write and an object API call are not the same thing. Putting a thin storage interface in your application early makes that migration an afternoon rather than a project.

Moving data between providers is the harder version of the same problem, mostly because of transfer charges and time, and moving data between cloud providers covers the mechanics. If you are still choosing where to put the data in the first place, provider pages such as Buy Scaleway Account and Buy UpCloud Account describe how each platform packages its storage products.

Related comparisons

AWS vs UpCloud
vs

AWS vs UpCloud

UpCloud sells storage performance and operational simplicity as the product. This comparison tests that against AWS for latency sensitive workloads…

AWS vs Google Cloud
vs

AWS vs Google Cloud

A comparison built around data gravity: how AWS and Google Cloud expect you to store, move and query data, and…

AWS vs IBM Cloud
vs

AWS vs IBM Cloud

IBM Cloud is built around OpenShift, regulated industries and hardware you cannot rent elsewhere. This comparison sets that platform thesis…

AWS vs Azure
vs

AWS vs Azure

Azure is usually chosen through identity and licensing rather than through infrastructure. This comparison covers enterprise agreements, hybrid estates, hierarchy…

Accounts discussed in this guide

Buy AWS Account
Buy AWS Account
Amazon Web Services

Buy AWS Account

Amazon Web Services has become synonymous with cloud computing itself, and for millions of developers and businesses around the world, an AWS account is…

From$25 14 options

Keep reading

IAM Fundamentals: Users, Roles and Least Privilege
IAM Fundamentals: Users, Roles and Least Privilege
Security and IAM 7 min read

IAM Fundamentals: Users, Roles and Least Privilege

Identity and access management decides who can do what in your cloud account. This guide covers users versus roles, how permission decisions…

Service Limits and Quotas Explained
Cloud Accounts 7 min read

Service Limits and Quotas Explained

Quotas decide what your account can create before you ever hit a technical ceiling. This guide covers soft and hard limits, where…

Need the account, not just the theory?

Browse documented configurations across every provider in the catalog, or ask us which one fits your workload.

Telegram