What is data durability?
Data durability is the probability that data you have stored stays complete and uncorrupted, and is never lost by the storage system. It is usually given as an annual percentage written in nines, such as 99.9999999%. A durability figure says how likely the data is to survive, not how likely you are to reach it at a given moment.
Think of a bank safe. Durability is the chance that your documents are still inside, undamaged, years from now. Availability is whether the bank is open when you walk up to the door. A closed bank is annoying, but your documents are still there.
What do the nines mean?
Each nine cuts the expected loss by a factor of ten. Durability of 99.9999999%, nine nines, means a stored object has roughly a one in a billion chance of being lost in a year.
Scale makes it concrete
Spread over many objects, the math gets easier to picture. If you store one billion objects at nine nines, you would expect to lose about one object a year. With one million objects, you would expect to lose one roughly every thousand years.
Where the figure comes from
Durability is a design figure, calculated from a model rather than measured. The model takes how often disks and servers fail, how fast the system rebuilds lost pieces, and how many pieces can be lost at once. Faster repair and more redundancy push the figure up.
How is durability different from availability?
Availability is the share of time you can read and write your data. A network outage or a server reboot makes data unavailable for a while, but nothing is lost, and access returns once the problem is fixed. Lost data, on the other hand, is unavailable forever.
The two numbers also live in different places. Availability is what an SLA usually covers, measured over a month: 99.9% allows about 43 minutes of downtime a month, or about 8.8 hours a year. Durability is usually stated for the design and is not part of the SLA.
What does durability not protect against?
A durability figure covers data loss caused by the storage system itself, such as failed disks, failed servers and silent corruption. Systems detect corruption with checksums and repair it from redundant pieces. Many common ways to lose data sit outside that figure:
- Deleting or overwriting an object by mistake, which the system carries out as asked.
- Ransomware or a leaked key used to delete or encrypt data.
- A fire, flood or power event that takes out the whole site.
- A bug in your own application that writes bad data.
Versioning, Object Lock and a second copy in another location cover these cases. Durability and backups solve different problems, so you need both.
How enum handles data durability
enum Object Storage erasure-codes every object with EC 4+2 across failure domains, and enum states 99.9999999% durability for this design. The 99.9% SLA covers availability, not durability. The design protects data within one zone, so use versioning, Object Lock and a copy elsewhere for data you cannot afford to lose.
FAQ
Is a durability figure a guarantee? No. It is a calculated design figure based on failure and repair rates. Durability is usually stated for the design, while the SLA covers availability, so check which number a contract actually covers.
Does higher durability mean faster or more reliable access? No. Durability says whether data survives, not whether you can reach it right now. A system can have very high durability and still be unreachable during a network or service outage.
Do I still need backups with nine nines of durability? Yes. Nine nines protects against failed hardware, not against accidental deletion, ransomware or losing the site. Keep versioned, locked copies and a second copy in another location for anything critical.