simulation: sim-realism: comment about ledger state impact

Author: Saizan

simulation/docs/sim-realism.tex

@@ -318,7 +318,8 @@ \section{Scenario \ref{sc:tcp}}
 
 
 A possibility is that IBs are not fully diffusing because nodes are unable to reconstruct the ledger state to validate them in.
-In Fig.~\ref{fig:scenario4-higher-IB-rate-send-recv-short-stage-no-cert} we try to see what happens when we require too many votes for EBs to certify, so the actual ledger state stays at Genesis. However IBs will still reference the most recent RB with a ledger state on the generating node, so other nodes will have to wait an see it (and not be on a differnet fork).\todo{try setting the reference RB as always Genesis} Even with that caveat we see IB diffusion improve, getting quite close to the one with 60 slot stage length. Ledger state reconstruction does seem a significant factor, since when EBs are 3 times rarer (Fig.~\ref{fig:scenario4-higher-IB-rate-send-recv}), or here where they do not get certified, diffusion does better.
+In Fig.~\ref{fig:scenario4-higher-IB-rate-send-recv-short-stage-no-cert} we try to see what happens when we require too many votes for EBs to certify, so the actual ledger state stays at Genesis. However IBs will still reference the most recent RB with a ledger state on the generating node, so other nodes will have to have adopted the same RB at some point.\todo{try setting the reference RB as always Genesis} Even with that caveat we see IB diffusion improve, getting quite close to the one with 60 slot stage length. Ledger state reconstruction does seem a significant factor, since when EBs are 3 times rarer (Fig.~\ref{fig:scenario4-higher-IB-rate-send-recv}), or here where they do not get certified, diffusion does better. Inspecting the logs we can further note that most IBs are validated immediately upon reception, meaning the ledger state was already available, and only a tiny fraction waits up to ~$3$s, meaning that the real impediment is that EBs can get certified even if a good percentage of nodes has not and will not validate some of the referenced IBs, because they lack the corresponding RB/ledger state.
+\todo{Keep ledger-state only for the preferred chain.}
 \scenarioVsIdealFigure{scenario4-higher-IB-rate-send-recv-short-stage-no-cert}{sc:tcp-higher-IB-rate-send-recv-short-stage-no-cert}
 
 Fig.~\ref{fig:scenario4-higher-IB-rate-send-recv-short-stage-no-cert-100-cpu} confirms the above with the bounded cpu worker pool, with a much larger than needed bound.