You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
in recent epoch there are many activesets that are slightly different. the root cause of this difference is suboptimal reconciliation
of known atxs before the start of the epoch. which should be addressed on its own.
however activesets can be encoded more efficiently as well. instead of generating new active sets smesher can lookup
previously generated activeset and extend/remove items from it based on local view.
builder selected an activeset and ready to publish reference ballot
pick activeset with the most weight from activesets received in this epoch
compute added/removed hashes and encode with reference to highest activeset
broadcast delta-activeset, if peer is missing a reference it should request it, but in most cases it should already have it as it was created earlier
use new protocol that relies on streaming.
note that peer can request data using previous (non-delta encoding) or a new delta-encoding for backward compatibility.
if peer requested non-delta encoding we should re-encode it after loading from database. eventually we will be able to drop non-delta encoded activesets, but we will need to maintain backward compatibility for several epochs.
both this and #5306 are aimed to tackle the same problem, but the latter one is more important, and we should do that first, maybe we won't even need to implement this one.
The text was updated successfully, but these errors were encountered:
)
related: #5282
this should reduce variability in encoded active sets.
previous grader was using hare preround delay, which is 25s and not sufficient for atx to propagate across the network.
)
related: #5282
this should reduce variability in encoded active sets.
previous grader was using hare preround delay, which is 25s and not sufficient for atx to propagate across the network.
in recent epoch there are many activesets that are slightly different. the root cause of this difference is suboptimal reconciliation
of known atxs before the start of the epoch. which should be addressed on its own.
however activesets can be encoded more efficiently as well. instead of generating new active sets smesher can lookup
previously generated activeset and extend/remove items from it based on local view.
note that peer can request data using previous (non-delta encoding) or a new delta-encoding for backward compatibility.
if peer requested non-delta encoding we should re-encode it after loading from database. eventually we will be able to drop non-delta encoded activesets, but we will need to maintain backward compatibility for several epochs.
both this and #5306 are aimed to tackle the same problem, but the latter one is more important, and we should do that first, maybe we won't even need to implement this one.
The text was updated successfully, but these errors were encountered: