PROVIDER - Distributed object storage

Storj

Storj erasure-codes objects into shards spread across many independent nodes and fronts them with an S3-compatible gateway, which is why it shows up as a backup and archive target billed on usage with no minimum commitment.

Backup & Migration United States 2026.11.20
3.6pts

Detailed Notes

Pros 5 items

An S3-compatible gateway with a standard endpoint means 1 line of configuration points existing S3 tools or backup software at Storj
Objects are erasure coded into 80 shards with 29 enough to rebuild, per vendor documentation, so a single node going offline does not block reads
No minimum storage volume and no minimum monthly fee, billed on the 2 meters of capacity used and egress traffic
Object Lock and versioning provide 2 tamper-resistance controls for immutable backup and compliance archive targets
Capacity is aggregated from independent storage nodes rather than 1 central facility, so growth follows node supply instead of a new data centre build-out

Cons 4 items

Latency and throughput depend on node topology, and random small-file IOPS trails a same-region centralized object store
There is no conventional region or availability zone concept, so compliance reviews need the distributed-storage model explained separately
Brand and ecosystem scale are small, and official certifications, plugins, and templates from mainstream backup vendors are limited
Access from mainland China relies on cross-border links, and large uploads and restores vary widely in speed

Service Overview

Storj is a distributed cloud storage vendor headquartered in the United States, and within Backup & Migration it rarely plays the lead role. It answers the question of where backup data finally lands: it provides object storage and fronts it with an S3-compatible gateway, so any backup tool that accepts a custom endpoint can point a repository at Storj.

The implementation differs from centralized object storage. On write, an object is erasure coded into many shards spread across storage nodes operated independently, and on read, enough shards are gathered to rebuild it; by vendor documentation each object splits into 80 shards of which 29 suffice. There is no single central facility to trust, and no single facility failure that takes everything down.

Billing is the reason it gets used heavily for backup and archive: no minimum storage volume, no minimum monthly fee, settled on actual capacity used and egress traffic. That suits cold data that is rarely read but must be retained for years, and it avoids the case of keeping 2 TB under compliance while paying for a 10 TB floor.

Key Strengths

  • Cheap S3-compatible onboarding: the gateway offers a standard endpoint and access keys, so existing S3 clients, backup tools, and lifecycle utilities switch with one configuration change and no code edit
  • No minimum commitment: billing follows capacity and egress, so archive workloads are not penalized by a floor and cost moves linearly with volume
  • Erasure-coded redundancy: shards spread across many independent nodes and a single node going offline stays recoverable, so fault tolerance does not rest on a dual-site design in one facility
  • Object Lock and versioning: both are supported, making it usable as the offsite or offline copy layer in a ransomware defence chain where locked copies cannot be overwritten
  • No capacity provisioning: nothing to pre-order and no region build-out to wait for, so cost follows writes and suits projects with uncertain growth

Product Ecosystem

Platform capabilities

The core capability is object storage itself: bucket management, multipart upload, presigned URLs, lifecycle rules, and access policies. The S3-compatible gateway translates requests into the internal protocol, so clients see a standard S3 API and aws-cli, rclone, restic, and similar tools work directly.

Adjacent capabilities include access grants, where API keys and access grants are separate objects, plus usage reporting and project isolation. Usage is metered per project across stored capacity, egress, and multipart request counts, which matters when several teams share one storage account; otherwise the bill shows only a total with no path back to a cost centre.

Integrations

  • S3 ecosystem: aws-cli, rclone, restic, and Duplicati all support custom endpoints and connect directly
  • Backup software: products with S3-compatible repository support can write through the gateway, with compatibility confirmed per product documentation
  • Custom applications: any service using an S3 SDK migrates by changing endpoint and credentials
  • Operations: access grants narrow permissions by bucket and prefix, and lifecycle rules expire stale objects automatically
  • Migration: bulk upload from other object stores or local directories, with multipart handling for large files

Typical scenarios

Three uses dominate: an offsite second copy of a backup repository, replicating local backups to satisfy 3-2-1; long-term archive, moving cold data out of expensive object storage for multi-year retention; and static assets or logs for applications, where S3 compatibility avoids code changes. All three share the profile of written once, read rarely, retained long.

Limitations

  • Distributed nodes bring latency variance, and random small-file performance trails a same-region centralized store, ruling out databases and high-frequency small-object workloads
  • With no region or availability zone model, compliance audits need a separate explanation of data placement and encryption boundaries
  • Ecosystem scale is small: official certifications and ready-made templates from mainstream backup vendors are limited, so compatibility needs testing in-house
  • Access from mainland China crosses long-haul links, and upload and restore rates fluctuate enough that recovery windows are hard to forecast precisely

Use Cases

  • Offsite copy of a backup repository (★★★★☆): S3 compatibility plus usage billing fits the offsite tier of 3-2-1, cheap to write and read only when needed
  • Cold data and compliance archive (★★★★): no minimum commitment means long-retained data is not charged for idle capacity
  • Projects with an existing S3 toolchain (★★★☆): rclone and restic only need a new endpoint, so migration cost approaches zero
  • High-frequency small files and online services (★★): latency and IOPS are not competitive, so it is a poor primary store for live services
  • Workloads centred on mainland China (★★): cross-border quality is inconsistent, so treat it as an offsite copy only, never the primary

Pricing

Item Price Notes
Stored capacity Billed per GB-month used No minimum volume commitment; unit price varies by tier and region, official quote governs
Egress traffic Billed per GB downloaded Restores and verification generate traffic, so frequent drills need separate estimation
Multipart and requests Billed per request Large files with multipart uploads generate far more requests than small objects
Object Lock and versioning No extra charge Retained versions increase capacity, so the cost lands in the storage line

The table lists billing dimensions only and excludes cross-border bandwidth and migration tooling. Compare on all-in cost per TB per year, weighting storage, egress, and requests by the real read-write ratio; a storage-only unit price understates restore-time traffic.

FAQ

Can Storj be used as backup software itself?

No, it is a storage target. Scheduling, incremental chains, and retention stay with the backup tool, whether that is restic, rclone, Duplicati, or a commercial product supporting S3 repositories. Storj supplies the object storage layer and S3-compatible interface.

Is data safe when it sits on other people's nodes?

Client-side encryption happens before writing, so shards carry no plaintext, and access needs both an API key and an access grant. Storage node operators therefore cannot read content, but self-built applications that skip client-side encryption must verify their own encryption chain rather than assuming the storage layer handles it.

Is restore speed guaranteed?

There is no hard bandwidth commitment in the SLA sense, and real throughput depends on node distribution and the cross-border path. Workloads sensitive to recovery time should keep a local or same-region copy and treat Storj as the offsite third copy rather than the only one.

How does it compare with mainland China object storage?

Keep the primary copy in-country for latency and compliance, and use Storj as the overseas offsite copy. Doing it the other way round creates both latency and compliance problems at once.

Code Examples

Point aws-cli at the Storj gateway

#!/usr/bin/env bash
set -euo pipefail

# Credentials come from environment variables; the endpoint is the S3-compatible gateway
export AWS_ACCESS_KEY_ID="${STORJ_ACCESS_KEY}"
export AWS_SECRET_ACCESS_KEY="${STORJ_SECRET_KEY}"

ENDPOINT="https://gateway.storjshare.io"

# List buckets and upload a test object to confirm the path works
aws s3 ls --endpoint-url "${ENDPOINT}"
aws s3 cp ./backup.tar.zst "s3://${STORJ_BUCKET}/nightly/backup.tar.zst" \
  --endpoint-url "${ENDPOINT}" \
  --checksum-algorithm CRC32

Large uploads switch to multipart automatically; part size and request count feed straight into the bill.

Upload with boto3 and verify

#!/usr/bin/env python3
# Upload a file, print the returned ETag, then list the prefix to confirm the object landed
import hashlib
import os
import boto3
from botocore.config import Config

session = boto3.session.Session()
s3 = session.client(
    "s3",
    endpoint_url=os.environ.get("STORJ_ENDPOINT", "https://gateway.storjshare.io"),
    aws_access_key_id=os.environ["STORJ_ACCESS_KEY"],
    aws_secret_access_key=os.environ["STORJ_SECRET_KEY"],
    config=Config(retries={"max_attempts": 5, "mode": "standard"}),
)

path = "backup.tar.zst"
key = f"nightly/{os.path.basename(path)}"

with open(path, "rb") as fh:
    local_md5 = hashlib.md5(fh.read()).hexdigest()

resp = s3.put_object(Bucket=os.environ["STORJ_BUCKET"], Key=key, Body=open(path, "rb"))
print("etag:", resp["ETag"], "local md5:", local_md5)

for obj in s3.list_objects_v2(Bucket=os.environ["STORJ_BUCKET"], Prefix="nightly/").get("Contents", []):
    print(obj["Key"], obj["Size"], obj["LastModified"])

For multipart uploads the ETag is no longer a plain MD5, so compare object size and manifest instead.

Community & Resources

Visit Storj Website → https://www.storj.io/
Visit Website ↗