Explaining group splits and merges

[dirvine]

@happybeing

So are there different kinds of group,

Yes, in fact there are different groups or Peers even in routing now (from our design talks).

• Infants :- Peers of age 4 or less, these do not affect churn (by extention age of other nodes, after network start).
• Adults :- Peers of age > 4 but not the oldest group_size peers in the section.
• Elders :- These are the oldest group_size peers in a section, these are the decision makers.

If you go back to my google talk in 2008 this was described. Then we have the peer groups in vaults as per the language of the network , this is possibly changing slightly with disjoint sections.

@happybeing

is the prefix used to determine which group looks after which chunk, based on the prefix of the hash of the chunk itself?

Yes, well it gives us the section the chunks will be in, likely now the elders will look after the chunk, i.e, be the DataManagers of the data, but the ManagedNodes will likely be the closest to the chunk as per that paper (it’s a small change but not flushed completely through yet (alpha 4). There may be small changes, but the responsibilities are still the same as they are easily calculable.

@Traktion

Would I be correct in asserting that the number of chunk copies direcrly correlates to group size or is there another dynamic at play? It sounds like data should be rather secure under regular operation, even without data chains (which is needed to cope with systemic system failures).

Number of chunks will be (in alpha 4) liely be 4. regardless of section size. Group size is a constant for security, section size is dynamic as @mav superbly highlighted and replicant size (or DataManager size is constant).

We will publish a load more info soon in a new RFC to tie data chains and disjoint sections together from a routing perspective for alpha 3, then for alpha 4 we will show how that works with data and communications. It sounds a bit complex but it is actually likely to be much less code with less time durations (local) and magic numbers, caches etc. So although difficult to reason about in many ways the end result should be very natural, simple and effective with a high degree of efficiency.