view slideshow.tex @ 8:60852a38c763 washlug

more stuff
author Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
date Thu, 19 Mar 2009 16:41:53 -0400
parents 6410a0499945
children 7a09d49f78ab
line wrap: on
line source

\documentclass{beamer}

\usepackage{beamerthemeshadow}

\title{XFS --- The Black Leather Jacket of filesystems}
\author{Josef ``Jeff'' Sipek\\ $<$jeffpc@josefsipek.net$>$}
\date{March 19, 2009}

\begin{document}
\frame{\titlepage}

%\AtBeginSection[]
%{
%\begin{frame}
%	\frametitle{Outline}
%	\tableofcontents[currentsection]
%\end{frame}
%}

\section{History}
\begin{frame}
	\frametitle{History: Stoneage (1992)}
	\begin{itemize}
		\item<2-> Berkeley's FFS was state of the art
		\item<3-> SGI's IRIX had EFS (FFS with extents)
		\item<4-> Many limitations
		\begin{itemize}
			\item<5-> Small file sizes (2 GB)
			\item<6-> Small filesystem sizes (8GB)
			\item<7-> Statically allocated metadata
			\item<8-> Long recovery times
			\item<9-> Slow operations on large directories
			\item<10-> No extended attributes
			\item<11-> No access control lists
		\end{itemize}
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{History: Enlightenment (1993)}
	\begin{itemize}
		\item<2-> SGI makes XFS
		\begin{itemize}
			\item<3-> 'X' being a variable
			\item<4-> To be replaced later
			\item<5-> Name stuck
		\end{itemize}
		\item<6-> Radically departs from traditional filesystems
	\end{itemize}
\end{frame}

%
% Features
%
\section{Features}
\begin{frame}
	\frametitle{Features: Allocation}
	\begin{itemize}
		\item<2-> Delayed allocation
		\begin{itemize}
			\item<3-> Allocate on flush
			\item<3-> Write reserves disk space, decide where later
		\end{itemize}
		\item<4-> Pre-allocation
		\begin{itemize}
			\item<5-> Reserve space before it is written
			\item<5-> Minimizes fragmentation
		\end{itemize}
		\item<6-> Direct I/O support
		\begin{itemize}
			\item<6-> Currently only Linux fs with parallel direct I/O
		\end{itemize}
		\item<7-> Stripe aware allocator
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Features: Inodes}
	\begin{itemize}
		\item<2-> Dynamically allocated in clusters
		\item<3-> Inode size is mkfs option
		\begin{itemize}
			\item<3-> 256 Byte (default) -- 2 kByte (max)
		\end{itemize}
		\item<4-> Extended Attributes
		\begin{itemize}
			\item<5-> $<$name, value$>$ pairs
			\item<6-> Used by ACLs, Capabilities, SELinux, DMAPI
		\end{itemize}
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Features: Other}
	\begin{itemize}
		\item<2-> B+Trees
		\item<3-> 64-bit
		\item<4-> Variable block size (512 Bytes -- 64 kB)
			\begin{itemize}
				\item 512 Bytes -- \texttt{\textbf{PAGE\_SIZE}}
			\end{itemize}
		\item<5-> Allocation Groups
		\item<6-> Lots of utilities
		\item<7-> Direct I/O
		\item<8-> DMAPI
		\item<9-> Extended Attributes/ACL
		\item<10-> Online fs growth
		\item<11-> Very fast repair time
		\item<12-> Amazingly fast
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Why not use ext2?}
	\begin{itemize}	
		\item<2-> Does not keep a journal
		\begin{itemize}
			\item<3-> fsck mandatory to recover after a crash
			\item<4-> fsck is dog slow
		\end{itemize}
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Why not use ext3?}
	\begin{itemize}
		\item<2-> 16TB filesystem limit
		\begin{itemize}
			\item<3-> Used to be 8TB until about 2 years ago
		\end{itemize}
		\item<4-> fsck is dog slow, although not as frequently
		needed as with ext2
	\end{itemize}
\end{frame}

\begin{frame}[fragile]
	\frametitle{Why not use ext4?}
	\begin{itemize}
		\item<2-> Not considered mature/stable
		\item<3-> \emph{Many} cases still not properly handled
		\item<4-> fsck is dog slow, although not as frequently
		needed as with ext2
		\item<5-> User-space still limited to 16TB
	\end{itemize}
	\uncover<6->{\begin{exampleblock}{}
This is the next generation of the ext3 filesystem.\\
...\\
If unsure, say N.
	\end{exampleblock}}
\end{frame}

\begin{frame}
	\frametitle{Why not use Reiserfs?}
	\begin{itemize}
		\item<2-> Not very actively maintained
		\item<3-> ``Forgotten''
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Why not use JFS?}
	\begin{itemize}
		\item<2-> Doesn't perform as well as it should
		\item<3-> Not as actively maintained as developed
		\item<4-> The maintainer contributes the odd fix
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Why not use btrfs?}
	\begin{itemize}
		\item<2-> Not stable/mature
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{Why \textit{use} XFS?}
	\begin{itemize}
		\item<2-> Stable, mature codebase
		\begin{itemize}
			\item<3-> Oldest journaling filesystem on Linux
		\end{itemize}
		\item<4-> Very well performing
		\item<5-> Very fast repair time (for the times you need it)
		\item<6-> DMAPI support (for HSM)
	\end{itemize}
%% That all sounds great, but is using XFS really worth it? Well, let me show
%% you an example... Once upon a time, there was a system admin.
\end{frame}

\begin{frame}
	\frametitle{Why \emph{not} use it?}
	\begin{itemize}
		\item<2-> No data journaling
		\item<3-> Got time to waste waiting for fsck to finish running
		\item<4-> Don't like leather jackets
	\end{itemize}
\end{frame}

%
% Demos
%
\newcommand{\demo}[1]{\begin{frame}\frametitle{#1}\begin{center}Demo\end{center}\end{frame}}

\section{Demos}
\demo{\texttt{\textbf{mkfs}} \& \texttt{\textbf{mount}}}
\demo{\texttt{\textbf{xfs\_check}} \& \texttt{\textbf{xfs\_repair}}}
\demo{\texttt{\textbf{xfs\_growfs}}}
\demo{\texttt{\textbf{xfsdump}} \& \texttt{\textbf{xfsrestore}}}
\demo{\texttt{\textbf{xfs\_bmap}}}
\demo{\texttt{\textbf{xfs\_io}}}
\demo{\texttt{\textbf{xfs\_quota}}}
\demo{\texttt{\textbf{xfs\_admin}}}
\demo{\texttt{\textbf{xfs\_freeze}}}

%
% Wrap-up
%
\section{}
\begin{frame}
	\frametitle{Questions?}
	\begin{center}
	\includegraphics[width=\textwidth]{xfs.eps}
	\end{center}
\end{frame}

\begin{frame}
	\frametitle{More info/Resources/Thanks}
	\begin{itemize}
		\item \url{http://oss.sgi.com/projects/xfs/}
		\item \url{http://www.xfs.org}
		\item \url{xfs@oss.sgi.com}
		\item \url{http://oss.sgi.com/projects/xfs/papers/ukuug2003.pdf}
		\item \url{http://en.wikipedia.org/wiki/Comparison_of_file_systems}
		\item Martin K. Petersen for the apt description of XFS
	\end{itemize}
\end{frame}

%
% Extra slides
%
\section{Extra stuff}
\begin{frame}
	\frametitle{ext4 faults}
	\begin{itemize}
		\item Where do I even \emph{begin}?!
		\item<2-> Does not compact extents
		\item<3-> ...
	\end{itemize}
\end{frame}

\begin{frame}
	\frametitle{What are extents?}
	\begin{itemize}
		\item<1-> Suppose you need to keep track of what blocks belong
		to a file
		\item<2-> Naturally: $<$offset, block nr$>$
	\end{itemize}
\end{frame}
\begin{frame} % should be continuation of previous frame
	\frametitle{What are extents?}
	(FIXME: graphic showing a file with blocks pointing to blocks on disk)
\end{frame}
\begin{frame} % should be continuation of previous frame
	\frametitle{What are extents?}
	\begin{itemize}
		% continue list from before
		\item<1-> All works well...
		\item<2-> ...unless you have millions of blocks (1 million 4K
		blocks is 4GB)
		\item<3-> Extents express a range of blocks instead of a single
		block
		\item<4-> Rely on the fact that many times, blocks are
		contiguously used within a file
		\item<5-> For example: $<$offset, nr blocks, block nr$>$
	\end{itemize}
\end{frame}
\begin{frame} % should be continuation of previous frame
	\frametitle{What are extents?}
	(FIXME: graphic showing a file with an extent instead of many
	pointers)
\end{frame}

\end{document}