48 lines
2.0 KiB
TeX
48 lines
2.0 KiB
TeX
%% SCRAP: archive/quality/regression/opp-02-report
|
|
%% SOURCE: docs/working/archive/quality/regression/opp-02-report.md
|
|
%% STATUS: HISTORICAL
|
|
%% FITS: none
|
|
%% EDITORIAL: lifted — prose rewritten to press voice
|
|
|
|
\section*{OPP~\#2: Rolling Window Width Optimization Report (November 2025)}
|
|
|
|
Optimization Opportunity~2 tested three rolling window width configurations
|
|
(2048, 4096, and 8192 entries) to identify the statistically optimal window
|
|
size for adaptive decay-slope inference. The experiment executed 180 runs
|
|
(60 per configuration) using the deterministic FORTH-79 test workload and the
|
|
regression detection framework established for the OPP series.
|
|
|
|
\subsection*{Results}
|
|
|
|
\begin{center}
|
|
\begin{tabular}{lccc}
|
|
\toprule
|
|
Configuration & Mean Time (ns) & vs.\ Baseline & Recommendation \\
|
|
\midrule
|
|
\texttt{WINDOW\_SIZE\_2048} & 5{,}432{,}109 & $-8.67\%$ & consider \\
|
|
\texttt{WINDOW\_SIZE\_4096} & 5{,}456{,}664 & $-8.26\%$ & not recommended \\
|
|
\texttt{WINDOW\_SIZE\_8192} & 5{,}310{,}504 & $-10.72\%$ & \textbf{lock} \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{center}
|
|
|
|
The OPP~\#1 baseline was 5{,}947{,}974~ns (decay slope 0.2, default window).
|
|
|
|
\subsection*{Regression Analysis}
|
|
|
|
No Type~1, 2, 4, 5, or 6 regressions were detected. All three configurations
|
|
produced Type~3 tail-latency warnings (P99 elevated 40--55\% above baseline);
|
|
analysis attributed this to increased inference computation during heartbeat
|
|
ticks at larger window sizes, not to a true regression in optimization quality.
|
|
The mean improvements (8--11\%) were judged to outweigh the P99 degradation.
|
|
\texttt{WINDOW\_SIZE\_4096} was Pareto-dominated and eliminated.
|
|
|
|
\subsection*{Lock Decision}
|
|
|
|
\texttt{ROLLING\_WINDOW\_SIZE = 8192} was locked for OPP~\#3 onward,
|
|
combining with the OPP~\#1 lock (\texttt{DECAY\_SLOPE\_Q48} $\approx 0.2$)
|
|
to form the parameter baseline for subsequent experiments. Cumulative
|
|
improvement at this point: $-10.72\%$ over the original baseline.
|
|
|
|
%% TODO(bob): Confirm whether OPP #3--5 results exist and where they are archived.
|