Data can be lost.
I put a few files on IPFS some time ago. That files are no longer accessible because I stopped hosting them.
> Site will never go down
As long as there is a node hosting the data you're looking for. So it can go down.
Data can be lost.
I put a few files on IPFS some time ago. That files are no longer accessible because I stopped hosting them.
> Site will never go down
As long as there is a node hosting the data you're looking for. So it can go down.
IPFS's own docs do have these additional higher-level guarantees — so they specifically will never go down.
One way to think of this is that, in a centralized model, the owner of an asset has to choose a single centralized CDN partner to collaborate with to distribute that asset, where the website you visit is really a pointer to the single (fallible) CDN partner.
Whereas, in a decentralized model, the owner of an asset can pick zero or more CDN partners who will all then "back" the asset on the DHT with their own mirrored copies of it. And I say zero or more, because the asset-owner can also still back the DHT asset with their own mirror on some personal node, without this causing much of a problem for them. (Where doing this wouldn't help in the centralized-CDN case, for two reasons: if the central CDN goes down, clients don't automatically switch to fetching from the origin that populated the CDN; and even if they did, a personal origin server would likely fall over under the traffic load. Whereas in IPFS, those nodes fetching copies — gateways et al — become (temporary) mirrors of the data themselves; so an origin server only ever has to serve O(log N) copies of an asset, not O(N).)
Also, in a decentralized model, besides the owner, anyone else who is interested in the asset's continued availability, can also choose to increase the data's availability in the network by mirroring it themselves (or by paying some pinning service to mirror it.)
Compare/contrast: organizations like universities who seed torrents of important data like Linux distro ISOs. IPFS is the BitTorrent model of data survivability/availability, but applied to websites / web assets.
----
Which is all the way one would present this until a few years ago, since IPFS used to be the forerunner in this space... but now there are decentralized protocols that are similar to IPFS but are "permanent" at the network level, e.g. https://www.arweave.org/. With these networks, as long as you pay the (one-time) costs related to writing data to the network, your assets will then be available on the network as long as the network itself exists.