Yes , we would use GROUP_SIZE and QUORUM as the values of N and P in this case.
2 Likes
So then considering the excellent reliability of the internet now-a-days is this N+P going to improve data reliability? Or is it mainly for the speed/security benefits rather than for data reconstruction?
I ask because years ago on NNTP “par2” was introduced, which is an implementation of N+P redundancy for binary files. In those days the posts to NNTP would see a number of errors in nearly all binary files. But I downloaded 100GB from NNTP (1st time in 6 months) and realised that my need for repairing files had dropped to nearly zero whereas years ago it was like 75%
5 Likes
With the plain copy setup, if you want to increase the number of vaults that would serve a read request (let’s ignore caching for a moment), you can’t do that without requiring more and more excess storage. With an N+P setup, you are free to adjust the vaults/chunk ratio without inflicting the network’s storage capacity.
The more pieces you have, the higher the chance that some of them will get lost, but the lower the cost to add another parity piece (in units of the original chunk size) and the higher the extra redundancy such a piece would bring (actually, I’m not sure that’s true; it doesn’t change much though); it’s a really good deal.
So, I can see two obvious benefits:
- chunks can be downloading from many-many more places at the same time (more useful for small files, or where the order by which the chunks are downloaded matters… though only for the first few chunks, or when local caching is not feasible)
- storage requirements can be closer to 100% (as opposed to 400%)
2 Likes