-
Notifications
You must be signed in to change notification settings - Fork 10
ZigZag DRG parents topology #153
Comments
Spawned from the discussion in #153 which concerned particularly the Chung's construction used to generate the parents from the expander graphs, this issue also discusses the generation of the base DRG parents. |
@schomatis Notice that the TR (pag.18) describes the encoding by "swaping" the vector x = (x_0, .. x_n-1) of labels in each layer. (they the pseudocode in the box at page 18). my question is: which approach is better for implementation? |
@irenegiacomelli I'm not sure I follow, the Anyway, to help me follow your reasoning, could you produce a simple example where the |
@schomatis My previous comments was about the fact that both these things are different in the TR compared to the paper (eg, the encoding in the TR uses the swiping, which is not used in the paper). |
I still don't quite see how that correction works, but I trust your analysis so feel free to close this issue if you think it's resolved. |
From Slack:
Closing this as resolved then, do reopen if new questions arise, let's continue the expansion (Chung) parents specific construction discussion in #144. |
We seem to have two conflicting sources of information to determine the topology of ZigZag graphs:
Paper Tight Proofs of Space and Replication which contains Figure 1.1 (and explanatory text in the "PoRep from ZigZag Expander DRGs" section).
Technical report (TR) Scaling Proof-of-Replication for Filecoin Mining which contains a pseudo-code definition of the
DAG.Parents
function (and also explanatory texts).Historically we have implemented what's described in the paper but lately more attention has been drawn to the TR because of the detailed pseudo-code. Even though both sources seem to indicate that the paper should be trusted over the TR (released before the paper) we would need an explicit confirmation of which source to rely on and a more comprehensive definition of how are ZigZag parents generated (ideally the figure and the pseudo-code should eventually converge to a definition accepted by all parties).
Discrepancies
Without going into the complete details, the parents obtained from the DRG and expander graphs, depending on the layer we're at, may either have their node indexes renumbered or their edge directions reversed (for succinctness we will associate in this description the verbs renumber and reverse with nodes and edges receptively). Renumbering and reversing (except for some corner cases) do not yield the same topology.
From the paper:
From the TR:
b = 1
case of the pseudo-code reveals comparisons of the renumberedj
index against the originali
one).The text was updated successfully, but these errors were encountered: