How permanent is data stored on IPFS?

From @randomshinichi on Sat Jan 30 2016 22:43:14 GMT+0000 (UTC)

In the whitepaper it says ‘Objects are permanent’.

How can they be if there is no guarantee that all the 256KB objects that make up my original file are on other people’s nodes? What if they ipfs repo gc? There has been talk about using IPFS as a versioned backup system but is my data really going to live forever out there?

Also, if I’m storing other people’s objects, then I could also read the data in them, right?

Copied from original issue:

1 Like

From @lgierth on Sat Jan 30 2016 22:52:37 GMT+0000 (UTC)

Permanence != Persistance

Permanence is about naming: content-addressing makes it so that an object’s name will always be the same. The “permanent web” is a web of links between objects with permanent names. The names are always the same, and thus the links won’t break.

Persistance is about storage: IPFS itself currently handles this by means of “pinning”, which excludes an object and its children from garbage collection within one IPFS node. In the future there will be more involved ways of managing persistance. One of them is Filecoin (paper), and there a couple of concrete ideas for an ipfs-cluster tool (ipfs/notes#58).

1 Like

From @lgierth on Mon Feb 01 2016 20:08:25 GMT+0000 (UTC)

This one is also related: Replication on IPFS – Or, the Backing-Up Content Model

From @jbenet on Wed Feb 03 2016 01:29:23 GMT+0000 (UTC)

To add to this, the way you should think about it is this:


  • read Replication on IPFS – Or, the Backing-Up Content Model
  • as long as anybody is willing to continue expending energy to maintain an object online, then that object is permanent. if nobody is willing to do that, then the object naturally cannot be transferred ad transferring requires energy expenditure.
  • this translates naturally to having both the traditional Web model, where publishers are also servers of the content.
  • but also includes that all readers share bandwidth costs (while caching)
  • and those readers who choose to keep the object too then become servers too, and as important as the original one.

Advanced Collaboration + Replication:

  • Now, this of course means that you can organize peers into networks that share the cost of backing each others’ content up, similar to how many other p2p systems work.
  • we are writing ipfs-cluster to improve how this will work.
  • in IPFS we have things like pincoop and ipfs-persistence-consortium – which aim to create collaborative pinning networks
  • But – for those who do not want to maintain their own infrastructure – the long term solution is to solve this with markets, where users can be at either end of the “resource contribution or resource consumption” spectrum, as will be seen in Filecoin and Ethereum.

From @RichardLitt on Tue Feb 28 2017 15:06:51 GMT+0000 (UTC)

Keeping this open so that it is more easily finable in the FAQ. Thanks. :slight_smile: