Limitations

Buckets

TypeLimitation
Bucket name3-63 characters long, case insensitive
Valid characters: 0-9, a-z and dashes (except at the beginning or end)
Bucket name cannot start with xn--
Bucket SizeUnlimited
Objects Per BucketUnlimited
Bucket count1.000 (soft limit, contact support to raise)
Anonymous AccessUnsupported, see S3 API Compatibility

Objects

TypeLimitation
Object Size5 TiB via multi-part upload
5 GiB via single part
Object Name1-1024 bytes of valid UTF-8
Multi-Part Uploads5 MiB - 5 GiB per part
1-10.000 parts per upload

Performance

The Intercolo Object Storage is optimized for capacity and throughput: storing large amounts of data reliably and transferring it at high speed. It is not optimized for very high request rates.

General performance expectations:

TypeExpectation
Write operationsAround 100 requests per second
Read operationsAround 1.000 requests per second
ThroughputMultiple Gbit/s, depending on object size (larger objects transfer more efficiently)

These are typical values, not hard limits. Actual performance is generallyhigher or can be lower, depending on object size, request mix, how many requests run in parallel and, as the object storage is a shared platform, the overall load from other customers.

Throughput scales with object size. Every request carries a fixed overhead for authentication, metadata and replication, which is negligible for an object of several megabytes but dominates for an object of a few kilobytes. A workload made up of large objects will therefore reach multiple Gbit/s comfortably, while the same amount of data spread across millions of tiny objects will be limited by the request rate long before it is limited by bandwidth.

Workloads that fit well

  • Backups, archives and database dumps
  • Media files, documents and other assets that are written once and read as needed
  • Build artifacts, container images and software releases
  • Log archives and data exports that are written in batches
  • Origin storage behind a CDN

Workloads that fit less well

The following workloads depend on very high request rates or very low latency and are not a good fit:

  • Very large numbers of small objects. Storing or reading many objects in the kilobyte range per second spends most of the time on per-request overhead.
  • Database-like access patterns. Using the object storage as a key-value store, a queue or the primary data store of a transactional application, with constant small reads and updates, which would typically be better served by a database or block storage.
  • Latency-critical requests. Response times are well suited to transferring data, but are not designed to stay in the single-digit millisecond range.
  • Serving end users directly at scale. For high volumes of public traffic, use our CDN service, which works in tandem with the object storage and takes the request load off your buckets.

Getting the most out of it

  • Combine small files into larger archives (for example tar or zip) before uploading, where your application allows it.
  • Use multi-part uploads for large objects and upload the parts in parallel.
  • Use a moderate number of parallel connections for large transfers instead of a single sequential stream.
  • Spread requests out evenly where possible, and make sure your client retries with a backoff on 503 SlowDown responses.
  • Use RenameObject instead of copying and deleting when moving objects, as it does not copy any data.

If you expect your workload to regularly exceed these figures, consists mainly of small objects, or you are unsure whether it is a good fit, please get in touch with support beforehand. We'll be happy to look at your use case and discuss what is possible.