\section{Introduction}

\subsection{Motivation}

In a previous work \cite{mreinhardt} developed the foundation of a new
network protocol, which is potentially beneficial for running parallel
scientific applications in cluster environments. This protocol showed
promising results in terms of latency reduction compared to TCP/IP,
but it lacked any provisions dealing with flow control or network
contention, it could not deal with these situations, and did not make
any progress with data transmission. This clearly shows, that these
facilities must not be omitted.

Therefore, the simplifications a solution tailored specifically for a
cluster environment may exploit, possibly lead to flow control scheme
which can maintain the benefit of the previous work. Such an solution
can possibly speed up the execution parallel applications while
retaining the main benefits of using standard ethernet hardwardware in
a cluster, which are the unrivalled price and the ubiquitousity.

\subsection{Task Formulation}

The objective of this diploma thesis is to get the results of a
previous work \cite{mreinhardt} ready for the use with scientific
applications while retaining its benefits of low CPU overhead and
latency \cite{mrei-cluster-perf}. One main problem is ESP's bad
performance in situations of network congestion.  Therefore a method
for flow control which fits the needs of a cluster environment shall
be developed and implemented.  A benchmark uncovering different
network congestion situations has to be implemented to show the
results of the optimization. Additionally, the CPU overhead has to be
assessed and compared with other implmementations (e.g. TCP/IP).
Finally, some selected applications shall be chosen to show the
benefits of the new approach compared to other techniques.

%%% Local Variables:
%%% mode: latex
%%% TeX-master: "main"
%%% End: