papers/lp-208-dag-mempool/sections/05-gc-and-throughput.tex

\section{Garbage Collection and Throughput Targets}

The DAG grows continuously. Without GC, validator memory and DHT
storage would grow without bound.

\subsection{Header GC}

A header at height $H$ is eligible for GC when:

\begin{itemize}[leftmargin=2em,nosep]
\item The consumer (LP-209) has total-ordered the DAG past height
  $H$, AND
\item The cert (LP-217 mode) covering all transactions hashed in the
  header has finalized.
\end{itemize}

Eligible headers are removed from the active DAG window. Their
content-hashes remain valid forever (anyone who archives a header
can still reference it), but live validators no longer hold them in
memory. Validator local storage of GC'd headers falls to ``cold''
-- moved to long-term archival storage; no longer indexed for active
parent-pointer resolution.

\paragraph{Default window.} $2 \times \textrm{cert\_timeout\_ms}$
worth of headers, plus the maximum LP-209 ordering-delay window.
Concrete reference: at $K{=}$\SI{50}{\milli\second}, LP-217
PQ-strict mode (cert $\sim$\SI{15}{\milli\second}), LP-209 Mysticeti
(ordering delay $\sim 3$ rounds) -- active window is
$5 \times $\SI{50}{\milli\second} $=$ \SI{250}{\milli\second} of
headers $=$ $5$ headers per validator $=$ $320$ headers in a
$64$-validator chain. Trivial memory.

\subsection{Body GC}

Bodies in the DHT (LP-201 layer~C) age out per the LP-201 TTL rule:
default \SI{24}{\hour} with refresh-on-access. Bodies for headers
that have been GC'd at all live validators are still served by
archival nodes until the TTL expires.

Bodies for headers still in the active DAG window are pinned (LP-201
content-pinning) so they cannot be evicted from the DHT.

\subsection{Restart / catch-up}

A validator that restarts re-fetches the active DAG window from
peers. Each peer serves its own freshest headers via the
unidirectional DAGHeader stream; the restarted validator stitches
the parent-pointers to reconstruct the DAG. Bodies are re-fetched
from the DHT as needed.

Catch-up latency: bounded by
$\textrm{active\_window\_size} \times \textrm{network\_RTT}$. For the
reference $320$-header window over a \SI{100}{\milli\second} RTT,
catch-up is $\sim 32$ seconds -- fast enough for routine validator
restart.

\subsection{Throughput targets}

The DAG mempool's per-validator emission rate sets the per-chain
throughput ceiling at the admission layer. Reference targets:

\begin{table}[h]
\centering
\small
\begin{tabular}{@{}rrr@{}}
\toprule
\textbf{Per-validator emit rate} & \textbf{64-validator chain} & \textbf{256-validator chain} \\
\midrule
\SI{10}{\kilo tx-hashes/s} & \SI{640}{\kilo tx/s} & \SI{2.56}{\mega tx/s} \\
\SI{50}{\kilo tx-hashes/s} & \SI{3.2}{\mega tx/s} & \SI{12.8}{\mega tx/s} \\
\SI{100}{\kilo tx-hashes/s} & \SI{6.4}{\mega tx/s} & \SI{25.6}{\mega tx/s} \\
\SI{1}{\mega tx-hashes/s} & \SI{64}{\mega tx/s} & \SI{256}{\mega tx/s} \\
\bottomrule
\end{tabular}
\caption{Per-validator emission rate $\times$ validator count $=$
chain-wide admission rate. \textbf{Projected.}}
\label{tab:throughput}
\end{table}

The per-validator emit rate is bounded by:

\begin{itemize}[leftmargin=2em,nosep]
\item \textbf{GPU verify} (LP-203). BLS leg-sig verify on Blackwell
  at $\sim$\SI{500}{\micro\second} aggregate per $N{=}64$; $\sim 2000$
  sig verifies/sec at full precision. For tx sig verify (per-tx, not
  per-header), GPU bench is $\sim$\SI{1}{\mega verifies/s} on
  Blackwell -- the per-validator emit rate at this hardware class is
  $\sim$\SI{1}{\mega tx-hashes/s}.
\item \textbf{DHT body write throughput} (LP-201). Bounded by
  replication factor $k{=}20$ and network bandwidth. At
  $\sim$\SI{1}{\kilo\byte} per tx and \SI{1}{\mega tx/s} emit, body
  throughput is $\sim$\SI{1}{\gibi\byte/\second} into the DHT per
  validator -- within Blackwell-class NIC bandwidth
  (\SI{200}{\giga\bit/\second} $\approx$
  \SI{25}{\gibi\byte/\second}).
\item \textbf{Mempool admission} (per-validator local). Bounded by
  tx-sig verify rate on the local GPU and Block-STM speculator
  throughput. Both scale with GPU class.
\end{itemize}

\paragraph{Honest reading.} The numbers above are projected from the
LP-203 measured bench rows extrapolated to the DAG emission model.
Measured DAG-mempool throughput will land in the LP-203 addendum
once the implementation tag ships.