4 ms·
As others have already pointed out, there are trust issues. The seller can fake the computation to return bogus result is one thing. But even if there is no m
by ezoe 9y ago
As others have already pointed out, there are trust issues.
The seller can fake the computation to return bogus result is one thing.
But even if there is no malicious intent, the resulting computation is still ended up bogus result, the malfunction of the hardware.
Commodity hardware isn't that reliable and there are so many commodity GPUs in the wild that looks like working but return incorrect result.
- alkonaut 9y agoall of these issues can be worked around by random validation, sumbitting all/some blocks to multiple people (redundant calculation). This is no different from existing efforts such as SETI@Homr and Folding@Home. The difference being the compensation which increases the incentive for fraud. The key to the trust issue is not trusting the remote computer, because you can’t. You have to validate somehow.
- ezoe 9y agoThat solution requires that all computation is deterministic and the result of computation can be easily comparable. I think not all the computation is that simple.
- alkonaut 9y agoIt only assumes that including incorrect results isn’t catastrophic for the end result, and that using mostly correct results will give a mostly correct outcome. If that isn’t the case I’d compute every block 2..n times instead.
- mathgeek 9y agoThis is an excellent point. You're also dealing with a market where demand is so high that GPU's running under constant high load are then resold for more than the MSRP of new ones.