46 lines
2.0 KiB
TeX
46 lines
2.0 KiB
TeX
\begin{figure*}[t]
|
|
|
|
\begin{subfigure}[t]{0.33\textwidth}
|
|
\includegraphics[width=\textwidth]{./graphs/activity/All_Devices__All_Apps__All_Queries__AllPreviousQueryCDFGraph.pdf}
|
|
\caption{}
|
|
\label{fig-overview-interarrival}
|
|
\end{subfigure}%
|
|
\begin{subfigure}[t]{0.33\textwidth}
|
|
\includegraphics[width=\textwidth]{./graphs/activity/All_Devices__All_Apps__All_Queries__AllRuntimeCDFGraph.pdf}
|
|
\caption{}
|
|
\label{fig-overview-runtime}
|
|
\end{subfigure}%
|
|
\begin{subfigure}[t]{0.33\textwidth}
|
|
\includegraphics[width=\textwidth]{./graphs/activity/All_Devices__All_Apps__All_Queries__AllRowcountCDFGraph.pdf}
|
|
\caption{}
|
|
\end{subfigure}%
|
|
|
|
\caption{\textbf{Summary Statistics for Android SQLite Queries.}}
|
|
|
|
\label{fig-overview}
|
|
|
|
\end{figure*}
|
|
|
|
Next, we look at overall workload characteristics of the queries observed
|
|
during our study. We examine how often queries arrive, how long they run, and
|
|
how many rows they return---all important inputs into desiging the TPC-Mobile
|
|
embedded database benchmark.
|
|
|
|
Figure~\ref{fig-overview} shows query interarrival times, run times, and
|
|
returned row counts (for \texttt{SELECT} statements) for all users,
|
|
applications, and non-informational query types (\texttt{SELECT},
|
|
\texttt{UPDATE}, \texttt{INSERT}, \texttt{DELETE}) included in our dataset.
|
|
Given that each mobile application is really generating an isolated workload
|
|
to its own embedded database, we measure query interarrival time only between
|
|
queries issued by the same application.
|
|
|
|
Examining the interarrival times shown in
|
|
Figure~\ref{fig-overview-interarrival}, it is interesting to observe that
|
|
many queries seem to arrive much more quickly than the minimum query runtime
|
|
shown in Figure~\ref{fig-overview-runtime}. Part of this may be due to apps
|
|
that use multiple separate databases, which is not yet captured by our
|
|
analysis. \XXXnote{However, our logging is also done above any locking
|
|
performed by SQLite, and so this may demonstrate that there are many cases
|
|
where multiple application threads are issuing overlapping queries in
|
|
parallel.}
|