\svnInfo $Id$  

\subsection{Experimental Setup}
To test the performance of \txcache, we used
\rubis~\cite{amza02:_specif_and_implem_of_dynam}, a auction website
prototype modeled after eBay. The benchmark simulates all of the
functionality of an auction site, such as browsing, bidding and
selling. The \rubis bidding benchmark consists of a mix of read-only
and read-write transactions with 15\% read-write transactions.

\begin{ednote}{IZ}
  Note the changes to \rubis and the things that we fixed like some of
  the results not being sorted and maybe some indices missing.
\end{ednote}

The experiments were run on several Dell PowerEdge SC1420 with 2GB of
RAM and a 3.2 GHz 2x Intel Xeon CPU. The database and web server each
ran on a single dedicated machine. The clients and cache servers ran
on separate machines, with a different number of machines dedicated to
each for different experiments.

\subsection{Modified Postgres Benchmarks}

In order to confirm that our modifications to the database do not
affect performance, we compared the performance of unmodified Postgres
to our modified Postgres. We ran the \rubis benchmark using the both
modified and unmodified databases with no external caching. Both
versions cpeaked at \XXX{} clients.


\subsection{\txcache Experiments}

The most important property of \memcached is its almost linear
scaling. 

\begin{figure}[tp]
  \centering
  
  \caption{\# of cache nodes vs. req/sec}
  \label{fig:txcache-scale-bench}
\end{figure}