To evaluate our solution, we checked the correctness of our results as well as we measured the throughput and latency of our system.
We made different experiments to determine latency and throughput of the queries. We run the experiments on an AMD FX 8350 with 4 GHz and 32 GB of main memory. The Java virtual machine for Odysseus Studio got 16 GB of main memory.
In the first experiments we wanted to determine what are the throughputs of the different queries. Figure~\ref{fig:throuhput} show the tuples that the systems processes per second for each query. The x-axis shows the corresponding play time (which we think is more informativ than the number of tuples processed until that time).

 \begin{figure}[h!]
  		\begin{center}
    		\includegraphics[width=.48\textwidth]{images/DatarateAllQueries.pdf}
    		\caption{Throughput in Tuple/sec}
    		\label{fig:throuhput}
    	\end{center}
\end{figure}


For that we initially measured how many tuple per seconds the system can process. Process means in this case: Read vom csv and write as csv. In the diagramm~\ref{fig:throuhput} this is the line Q0.