next up previous contents
Next: 6.2.4 Host Selection Up: 6.2.3 Load Polling Previous: 6.2.3 Load Polling

6.2.3.1 Estimation of the Effective System Load

SIESTA uses as much knowledge about the state of a host as possible in order to overcome the dilemma of a delayed value of the operating system load. The diagram depicted in Figure 6.4 shows how SIESTA deals with this issue in order to obtain an acceptable value for the effective load of a computer which is feasible as a basis for load control leading to a stable system.
\begin{Figure} % latex2html id marker 6488\centering \includegraphics{fig/virt... ...corrects the delayed load that is reported by the operating system} \end{Figure}

SIESTA introduces a virtual host which encapsulates the actual host and instead of communicating with the real host it does its business with the hosts virtual substitute. Since this virtual host knows everything about jobs that are running on itself, it is able to correct that portion of the host's load which was produced by jobs that have been submitted by SIESTA. For each of these jobs the virtual host adds a correction term which compensates the delay inherent to the operating system load. Figure 6.5 shows this procedure in detail.
\begin{Figure} % latex2html id marker 6496\centering \includegraphics[width=0.... ...ected by $l_{corr}$\ in order to get its effective load $l_{eff}$.} \end{Figure}
los is the load which is reported by the operating system, whereas lcorr denotes the correction which is added in order to obtain the effective load leff.

\begin{displaymath}l_{eff}= l_{os} + \underbrace{ \underbrace{\sum_{i=1}^N e^{... ...{stop}}{\tau}}\right]} _{\mbox{finished jobs}} }_{l_{corr}} \end{displaymath}

next up previous contents
Next: 6.2.4 Host Selection Up: 6.2.3 Load Polling Previous: 6.2.3 Load Polling
Rudi Strasser
1999-05-27