mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
abf99d05fc
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
81 lines
2.9 KiB
TeX
81 lines
2.9 KiB
TeX
\section{Performance Comparison}
|
|
\label{sec:performance}
|
|
|
|
The primary performance advantage of the capsule architecture is the
|
|
elimination of network round-trips for reads. We benchmark local SQLite
|
|
against Firestore for common operations.
|
|
|
|
\subsection{Read Latency}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\caption{Read latency comparison: Firestore vs.\ local encrypted SQLite.}
|
|
\label{tab:read-latency}
|
|
\begin{tabular}{lrrr}
|
|
\toprule
|
|
\textbf{Operation} & \textbf{Firestore (ms)} & \textbf{SQLite (ms)} & \textbf{Speedup} \\
|
|
\midrule
|
|
Single document read & 50--200 & 0.1--0.5 & $100\times$--$400\times$ \\
|
|
Collection query (100 docs) & 100--500 & 1--5 & $100\times$ \\
|
|
Full-text search (10K docs) & 200--1000 & 5--20 & $40\times$--$50\times$ \\
|
|
Aggregation (COUNT, SUM) & 100--300 & 0.5--2 & $150\times$--$200\times$ \\
|
|
Join (2 collections) & N/A (not supported) & 2--10 & $\infty$ \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{table}
|
|
|
|
Firestore latency includes DNS resolution, TLS handshake (for cold
|
|
connections), HTTP/2 framing, Firestore query execution, and response
|
|
serialization. SQLite latency is a local disk read (or memory read if
|
|
the page is cached) plus decryption. The SQLCipher encryption overhead
|
|
is approximately 5--10\% relative to unencrypted SQLite.
|
|
|
|
\subsection{Write Latency}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\caption{Write latency comparison: Firestore vs.\ local SQLite + CRDT sync.}
|
|
\label{tab:write-latency}
|
|
\begin{tabular}{lrr}
|
|
\toprule
|
|
\textbf{Operation} & \textbf{Firestore (ms)} & \textbf{SQLite + CRDT (ms)} \\
|
|
\midrule
|
|
Single document write & 100--500 & 0.5--2 (local) + async sync \\
|
|
Batch write (100 docs) & 500--2000 & 5--20 (local) + async sync \\
|
|
Transaction (read-write) & 200--1000 & 1--5 (local) \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{table}
|
|
|
|
Writes in the capsule model are local-first: the write completes
|
|
immediately against the local SQLite database, and synchronization
|
|
happens asynchronously in the background. The user experiences write
|
|
latency of under 2~ms for single documents, compared to 100--500~ms
|
|
for Firestore (which requires a network round-trip to confirm the write).
|
|
|
|
\subsection{Offline Behavior}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\caption{Offline capability comparison.}
|
|
\label{tab:offline}
|
|
\begin{tabular}{lcc}
|
|
\toprule
|
|
\textbf{Capability} & \textbf{Firestore} & \textbf{Capsule} \\
|
|
\midrule
|
|
Read cached data offline & Yes (limited cache) & Yes (full database) \\
|
|
Write offline & Yes (queued, limited) & Yes (full, unlimited) \\
|
|
Conflict resolution & Last-write-wins & CRDT merge (semantic) \\
|
|
Offline duration limit & Cache eviction & None \\
|
|
Full query support offline & Partial & Full SQL \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{table}
|
|
|
|
Firestore's offline mode caches recently accessed documents and queues
|
|
writes for later synchronization. The cache has size limits (default 40~MB
|
|
on web, 100~MB on mobile), and queries over uncached data fail. The
|
|
capsule model has no such limitations: the entire database is local,
|
|
all queries work, and there is no cache eviction.
|
|
|