What Is Erasure Coding? How It Compares to Replication

Erasure coding is a data protection method that splits a file into n fragments, any k of which can rebuild it. The system survives losing up to n-k fragments while storing far less than full copies would. AIOZ Storage applies Reed-Solomon erasure coding to every object you upload.

TL;DR:

• Erasure coding splits a file into n fragments where any k of them rebuild the original, so you can lose up to n-k fragments without losing the file.
• It needs much less storage than replication for the same durability. Storj's documentation notes that 10 full copies still can't beat a k=16, n=32 code, which only doubles the storage.
• Storj, Sia and AIOZ Storage all use erasure coding as their default redundancy. AIOZ Storage uses Reed-Solomon coding, and no single node holds a complete copy of your object.

Updated Oct 19, 2026: AIOZ Storage protects every object with Reed-Solomon erasure coding, and the opening, TL;DR, AIOZ section and FAQ now explain how.

What Is Erasure Coding?

Erasure coding encodes a block of data into n fragments, where any k of them are enough to rebuild the original. It doesn't matter which k you have. The two numbers define the scheme: k is the minimum you need to recover the file, and n is the total spread across storage nodes. Lose up to n-k fragments to failed nodes, and you can still read the file.

The math is the same family as Reed-Solomon codes, which CDs and DVDs use to survive scratches. Here the "scratch" is a node going offline instead of a damaged disc.

The storage cost of a scheme is its expansion factor, n divided by k. Say you store a 10 MB file with k=10 and n=20. That's an expansion factor of 2, so the file becomes 20 fragments of about 1 MB each, 20 MB in total. Scale both numbers up to k=20, n=40 and the expansion factor stays at 2. Durability improves, though, because more fragments now have to fail before you lose the file.

How Does Erasure Coding Compare to Replication?

Replication stores complete, separate copies of a file on several nodes. It's simpler, and it costs much more storage for the same durability. Storj's documentation puts it directly: "replicating data at 10x can't beat erasure codes with k=16, n=32, which is an expansion factor of only two."

So 10 full copies, 10 times the storage, still fall short of a code that costs twice the file's size. The reason is granularity. Replication survives only while at least one whole copy remains intact. Erasure coding survives any combination of lost fragments up to its threshold.

Replication still has its place. Each copy can serve a read on its own, and you don't need to decode anything to rebuild a file. Which approach a network picks depends on whether it optimizes for disk cost or for simplicity.

Which Decentralized Storage Networks Use Erasure Coding?

Storj, a decentralized S3-compatible network, uses erasure coding as its core mechanism. Its documentation gives the example that you need "29 of the 80" pieces to retrieve a file. When the healthy pieces of a segment drop to a repair threshold, Storj's repair process re-encodes the segment. It regenerates the missing pieces and uploads them to healthy nodes.

Sia uses Reed-Solomon erasure coding with a 10-of-30 scheme. Its documentation says files are divided into 30 segments and "any 10 of 30 segments can fully recover a user's file." That means up to 20 of the 30 hosts can go offline at once without data loss.

Filecoin works differently. Its documentation describes erasure coding as something a client can request when making a storage deal, not as an automatic default. Its main protocol-level redundancy is replication: the client stores full copies with several independent storage providers. So if you compare networks, don't assume every one defaults to erasure coding.

AIOZ Storage uses Reed-Solomon erasure coding on every object by default. It doesn't publish its k and n values, so the next section describes the mechanism instead of the numbers.

How AIOZ Storage Handles It

AIOZ Storage protects every object with Reed-Solomon erasure coding, per AIOZ Storage engineering (2026). After server-side encryption, each object is split into data pieces plus parity pieces, n in total, and the pieces go to different nodes on AIOZ DePIN. Any k of those n pieces rebuild the object, so it survives the loss of several pieces at once. No single node holds a complete copy of your object.

Encryption comes first. The docs say uploaded content is split into segments, each encrypted with its own random key (How Encryption Is Implemented). The coding then runs on that encrypted data, so each node stores pieces of ciphertext. The network also audits the nodes that hold pieces and repairs lost ones, the same job Storj's repair process does.

You don't configure any of this. Uploads use the standard S3 call from the docs, with two client settings first:

export AWS_REQUEST_CHECKSUM_CALCULATION=when_required
aws configure set default.s3.addressing_style path
aws s3 --endpoint-url https://s3.aiozstorage.network cp /path/to/file s3://bucket-name

The first two lines turn off the aws-chunked uploads and bucket subdomains that current AWS CLI versions default to, because AIOZ Storage needs a Content-Length on uploads and path-style URLs (checked Oct 19, 2026).

What we don't publish yet: the k and n values and a durability figure. The expansion-factor math above uses Storj's and Sia's published schemes for that reason.

To try it, follow the AWS CLI guide in the AIOZ Storage docs.

Frequently Asked Questions

What do "k" and "n" mean in erasure coding?
n is the total number of fragments a file is split into. k is the minimum number of those fragments needed to rebuild it. Any k of the n fragments work.

How much more storage does erasure coding need than the original file?
Its expansion factor, n divided by k. A 29-of-80 scheme, like Storj's example, has an expansion factor of about 2.76, so the fragments take roughly 2.76 times the file's size.

Is erasure coding always cheaper than replication?
For the same durability, it uses much less storage. Storj's documentation states that 10x replication still doesn't match an erasure code with only a 2x expansion factor.

Which decentralized storage networks use erasure coding?
Storj (29 of 80 pieces) and Sia (10 of 30) run it as their default, per each project's documentation. AIOZ Storage uses Reed-Solomon erasure coding too, without published k and n values.

Does Filecoin use erasure coding?
A client can request it for a storage deal, but Filecoin's default redundancy is replication across several storage providers.

Does AIOZ Storage use erasure coding?
Yes. Each object is encrypted, then Reed-Solomon coded into data and parity pieces on different AIOZ DePIN nodes. Any k of the n pieces rebuild it, and no node holds a complete copy.

Is replication worse than erasure coding?
Not worse, just a different tradeoff. It's simpler to run and read from, and it costs more storage for the same durability.

Related Reading

• What Is Data Replication? How Distributed Storage Systems Stay Available - the full-copy approach that erasure coding is measured against, with numbers from HDFS and Cassandra.
• AIOZ Storage Explained: S3-Compatible Decentralized Storage - what AIOZ Storage is overall, including how it protects the objects you store.
• What Is Amazon S3? Object Storage Explained - the object storage model this article's redundancy discussion sits on top of.
• AIOZ Storage vs Storj: Two Decentralized S3-Compatible Platforms Compared - two erasure-coded DePIN networks compared on pricing, access grants and encryption.
• Data Durability vs Availability: What 11 Nines Means - the two guarantees erasure coding and replication are both trying to earn, with AWS and Google Cloud's own published numbers.

References

• Storj Documentation: Understanding File Redundancy, Durability, Expansion Factors, and Erasure Codes
• Storj Documentation: Understanding File Repair in a Distributed Network
• AIOZ Storage Docs: How Encryption Is Implemented
• AIOZ Storage Docs: AWS CLI
• Sia Documentation: Is My Data Secure?
• Filecoin: How Storage and Retrieval Deals Work