2106 \subsection{Timing}

  2107 \label{sec:usertiming}

  2108 

  2109 An interesting aspect is the timing of the userspace library calls compared to

  2110 those of the kernel API. Table~\ref{tab:usertiming} shows the call times and

  2111 standard deviancies of typical (and time-critical) API functions measured on

  2112 an Intel Pentium 4 M CPU with \unit{2.2}{\giga\hertz} and a standard 2.6.26

  2113 kernel.

  2114 

  2115 \begin{table}[htbp]

  2116   \centering

  2117   \caption{Application Interface Timing Comparison}

  2118   \label{tab:usertiming}

  2119   \vspace{2mm}

  2120   \begin{tabular}{l|c|c|c|c}

  2121 

  2122     &

  2123     \multicolumn{2}{|c}{\textbf{Kernelspace}} &

  2124     \multicolumn{2}{|c}{\textbf{Userspace}}  \\

  2125 

  2126     \textbf{Function} &

  2127     $\mu(t)$ &

  2128     $\sigma(t)$ &

  2129     $\mu(t)$ &

  2130     $\sigma(t)$ \\

  2131     \hline

  2132 

  2133     \lstinline+ecrt_master_receive()+ &

  2134     \unit{1.1}{\micro\second} &

  2135     \unit{0.3}{\micro\second} &

  2136     \unit{2.2}{\micro\second} &

  2137     \unit{0.5}{\micro\second} \\

  2138 

  2139     \lstinline+ecrt_domain_process()+ &

  2140     \unit{<0.1}{\micro\second} &

  2141     \unit{<0.1}{\micro\second} &

  2142     \unit{1.0}{\micro\second} &

  2143     \unit{0.2}{\micro\second} \\

  2144 

  2145     \lstinline+ecrt_domain_queue()+ &

  2146     \unit{<0.1}{\micro\second} &

  2147     \unit{<0.1}{\micro\second} &

  2148     \unit{1.0}{\micro\second} &

  2149     \unit{0.1}{\micro\second} \\

  2150 

  2151     \lstinline+ecrt_master_send()+ &

  2152     \unit{1.8}{\micro\second} &

  2153     \unit{0.2}{\micro\second} &

  2154     \unit{2.5}{\micro\second} &

  2155     \unit{0.5}{\micro\second} \\

  2156 

  2157   \end{tabular}

  2158 \end{table}

  2159 

  2160 The test results show, that for this configuration, the userspace API adds

  2161 about \unit{1}{\micro\second} delay for each function.

  2162