garage-unix-socket/doc/talks/2021-04-28_spirals-team/introduction.tex

99 lines
2.8 KiB
TeX
Raw Normal View History

\section{Introduction}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}{A very casual motivation}
\begin{center}
\Large
I want to host \textbf{resilient web services} with \textbf{acceptable performance} on commodity hardware behind \textbf{household networks}.
\end{center}
\vfill
\begin{block}{Keywords}
\begin{columns}
\column{.5\columnwidth}
\begin{itemize}
\item Decentralised networks
\item Edge computing
\end{itemize}
\column{.5\columnwidth}
\begin{itemize}
\item Distributed storage
\item Privacy
\end{itemize}
\end{columns}
\end{block}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}{Context}
\textbf{Resilience}: Ability to recover quickly from failures and changes.
\vspace{1ex}
Only achievable through distribution of the hosted applications across several physical locations.
\vfill
\begin{block}{Application = \textbf{computations} on \textbf{data}}
\begin{itemize}
\item \textbf{Computation}: Stateless; easy to distribute \& orchestrate.
% where it is performed does not matter as long as the application's state is accessible R/W
%Computation units are "easy" to distribute and orchestrate.
\item \textbf{Data}: Stateful; hard to distribute \& full of trade-offs.
\end{itemize}
\end{block}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}{Concurrent writes example}{How to lose vaccines}
\centering
\includegraphics[width=.5\columnwidth]{figures/conflict_problem.pdf}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}{The problem}
% \textbf{Path dependency}: existing data stores are built for data centres.
% $\implies$ They assume good inter-node connectivity.
% \vfill
\begin{center}
\Large
Can we design an available data store tailored for adverse network conditions?
\end{center}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Maybe more framing of the context. What kind of data storage? Object vs Block vs what?
% \begin{frame}{``Stateless'', ``serverless'', and the elephant in the room}
% It seems easy to deploy \& administer web services nowadays ...
% Because the inherent complexity is shadowed by proprietary ``cloud'' solutions.
% The IT crowd can gloss over ``statelessness'' \emph{ad nauseam} ...
% But storing \emph{state} remains an open research problem.
% Data storage is either:
% \begin{itemize}
% \item A single point of failure;
% \item Delegated to proprietary solutions;
% \item Pain.
% \end{itemize}
% Today, we will review networked storage's history, and discuss open research questions.
% \end{frame}