emagdnim2100

> How does this system deal with the "data withholding" problem? In other words, when people provide "storage power" their data will be repeatedly sampled to make sure it is available... but when an entity claims that samples aren't being provided as required by the protocol, how does the system determine that that person wasn't lying, if the sampled data is still provided correctly in a followup request?

Filecoin sort of splits this problem into two parts – "data withholding" from Filecoin's proof-of-spacetime consensus mechanism (a "storage fault" in Filecoin terminology, yes I know there's a lot of new terminology here!), and "data withholding" from a client that's requesting stored data.

Storage miners are required to prove to the network itself, not to any specific challenger entity, that they're storing files. Each storage miner is (basically) randomly challenged once per [short interval] to provide a compressed cryptographic proof in response to a challenge. The proof conclusively confirms that, during that period, the miner was storing the data being they'd previously promised to store. You can ctrl-f "if a miner goes offline" in the linked post for a surface-level description of how the network deals with storage faults. Ditching the data and recovering it later is economically irrational for pretty involved reasons – basically, recovery is more expensive than just storing the data over the short-ish intervals during which faults are recoverable.

When it comes to "withholding data" from clients – retrieval on Filecoin is just a market-based system for bandwidth. The solution to holding data "hostage," i.e. refusing to serve it at reasonable prices, is to store a few replicated copies (just like centralized storage services do for you today behind the scenes). There's really no upside to miners refusing to profitably serve you a file when they know or suspect you can get it from another source.

> The "verified clients" are certified by "a decentralized network of verifiers". How does this system prevent a sibyl attack, i.e how does it prevent verifiers from repeatedly verifying themselves using multiple accounts?

The short answer here – with apologies for the brevity; details forthcoming – is that verified data isn't meant to be scarce, and some degree of over-verification is expected. There will be a decentralized group of folks responsible for (quite permissively) verifying and renewing clients for fixed amounts of data, and declining to renew allocations for clients who seem to be abusively verifying data. We're optimistic that this will dramatically decrease the rate at which "fake" data is stored and (most importantly) ensure that there's always storage available for client data.

> Why is it that erasure coding isn't necessary in this system?