Glossary
Petabyte storage
Petabyte storage is a single storage system holding 1,000 terabytes or more, a size where methods built for smaller arrays stop working. Drive failure becomes a routine event, so protection has to be efficient and repair has to happen automatically.
When failure becomes ordinary
A small array loses a drive occasionally and an administrator replaces it. A petabyte system contains hundreds or thousands of drives. At annual failure rates of one to two percent, as reported in published fleet statistics, failures arrive on a regular basis instead of as incidents. The design question changes from how to respond to a failure to how to keep a failure from needing a response at all.
How long does a rebuild take?
Rebuild time sets reliability, and drive sizes have raised it. A 32 TB drive rewritten at 150 MB/s takes about 59 hours, a little under two and a half days. In a traditional array that rebuild runs through one replacement drive, and the group is exposed to a second failure for the whole of it. As drive capacity grows faster than drive speed, that window keeps lengthening.
Why do RAID and triple copies fall short?
| Approach | What it costs at this size | Limit |
|---|---|---|
| RAID groups | Rebuild funnels into one replacement drive and runs for days | Long exposure window as drives grow |
| Three full copies | Three times the raw capacity | Power, space and drive cost multiply at petabyte scale |
| Erasure coding with distributed rebuild | Parity overhead well under that of copies; many servers share the repair | More processing and network traffic during writes and repairs |
The combination used at this scale is erasure coding with repair spread across many nodes, so each server rebuilds a small slice at once and the exposure window shortens as the system grows. That ties directly to data durability, which depends on how fast lost redundancy is restored.
Growth and operations
A petabyte is reached and passed, not bought once. Scale-out storage adds nodes as capacity is needed and moves data off old ones, so growth and hardware refresh do not need a migration project. At this size the storage also usually carries mixed workloads, such as backup and archive alongside analytics, which raises the importance of storage QoS so that one workload cannot crowd out another.
Frequently asked questions
How many drives make a petabyte?
With 20 TB drives, 50 drives hold a raw petabyte. After erasure coding overhead and spare capacity, a usable petabyte needs somewhat more depending on the layout.
Is petabyte storage always object storage?
Not always, but most new petabyte deployments for unstructured data use object storage because it scales by adding nodes and uses erasure coding natively. File systems at this size exist, mainly for high-performance computing.
What grows fastest in a petabyte system?
Unstructured data: backups, media, logs, images and analytics inputs. These are typically written once and read rarely, which makes protection efficiency matter more than raw speed.
Related terms
- Exabyte: The next scale step, where data cannot be moved at all.
- Erasure coding: How protection stays efficient at scale.
- Scale-out storage: Growing by adding nodes.
- Data durability: What failure rates and rebuild speed produce.
- Object storage: The usual platform for petabyte unstructured data.