Most Xen related slides except for xSplice

diff --git a/talk.pdf b/talk.pdf
index aecaada6f18b762de5e3436bc9230bf6dfbbd642..52e803c6872881dcf4486fc1ccc707b87012ac0f 100644 (file)
Binary files a/talk.pdf and b/talk.pdf differ

diff --git a/talk.tex b/talk.tex
index bff4298d04bdeb013047ebf6414f123c0a3f4159..f07a58201eb9adcd6e6ce2c718a4da31510254e7 100644 (file)
--- a/talk.tex
+++ b/talk.tex
@@ -237,6 +237,275 @@
     \end{itemize}
 }
 
+\section{Xen}
+
+\begin{frame}
+    \frametitle{The full virtualization spectrum}
+    {\centering
+    \begin{columns}
+        \begin{column}{0.4\textwidth}
+            {\scriptsize
+            \begin{table}
+            \centering
+            \tabulinesep=1.2mm
+            \begin{tabu} {| c | l}
+            \cline{1-1}
+            VS & Software virtualization \\
+            \cline{1-1}
+            VH & Hardware virtualization \\
+            \cline{1-1}
+            PV & Paravirtualized \\
+            \cline{1-1}
+            \end{tabu}
+            \end{table}}
+        \end{column}
+        \begin{column}{0.4\textwidth}
+            {\scriptsize
+            \begin{table}
+            \centering
+            \tabulinesep=1.2mm
+            \begin{tabu} {| c | l}
+            \hhline{-~}
+            \cellcolor{red!80} & Poor performance \\
+            \hhline{-~}
+            \cellcolor{yellow!80} & Room for improvement \\
+            \hhline{-~}
+            \cellcolor{green!80} & Optimal performance \\
+            \hhline{-~}
+            \end{tabu}
+            \end{table}}
+        \end{column}
+    \end{columns}}
+    \vspace{-2em}
+    {\scriptsize
+    \begin{table}
+    \centering
+    \tabulinesep=1.2mm
+    \begin{tabu} {| l | c | c | c | c | c |}
+    \multicolumn{1}{r}{}
+    & \multicolumn{1}{c}{\rot{Disk and network}}
+    & \multicolumn{1}{c}{\rot{Interrupts and timers}}
+    & \multicolumn{1}{c}{\rot{Emulated motherboard}}
+    & \multicolumn{1}{c}{\rot{\specialcell{Privileged instructions\\and page tables}}}
+    \\
+    \hline
+        HVM &
+        \cellcolor{red!80}VS &
+        \cellcolor{red!80}VS &
+        \cellcolor{yellow!80}VS &
+        \cellcolor{green!80}VH \\
+    \hline
+        HVM with PV drivers &
+        \cellcolor{green!80}PV &
+        \cellcolor{red!80}VS &
+        \cellcolor{yellow!80}VS &
+        \cellcolor{green!80}VH \\
+    \hline
+        PVHVM &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{yellow!80}VS &
+        \cellcolor{green!80}VH \\
+    \hline
+        PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{yellow!80}PV \\
+    \hline
+        PVH &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}PV &
+        \cellcolor{green!80}VH \\
+    \hline
+    \end{tabu}
+    \end{table}}
+\end{frame}
+\note{
+    \begin{itemize}
+        \item PV and HVM at both ends
+        \item To provide better IO performance in HVM guests, two new modes have been added
+        \item HVM with PV drivers is a fully virtualized domain with PV drivers for both disk and network devices, providing better performance (example: Wind
+ows, FreeBSD)
+        \item PVHVM is another step in the HVM with PV drivers, providing also PV drivers for interrupts and timers (example: Linux)
+        \item This improvements provide better performance, but we still need to run Qemu in the host to emulate a motherboard and the boot sequence.
+    \end{itemize}
+}
+
+\begin{frame}
+    \frametitle{Why PVH?}
+    \begin{itemize}
+        \item Performance: use hardware feature as much as possible
+       \item Security
+       \begin{itemize}
+               \item No emulation eliminate a main class of security bugs
+               \item No PVMMU etc, a lot less complex code for both guest kernel and Xen toolstack
+       \end{itemize}
+       \item Maintanence
+       \begin{itemize}
+               \item No PVMMU etc, a lot less code
+       \end{itemize}
+    \end{itemize}
+\end{frame}
+\note{
+    \begin{itemize}
+    \end{itemize}
+}
+
+
+\begin{frame}
+    \frametitle{Gory details about PVH}
+    \begin{itemize}
+        \item PVH-classic vs HVMlite-nodm
+       \item PVH-classic is first attempt for the design, to make PV guest look like HVM guest
+       \item HVMlite-nodm is the new approach, to make HVM guest look like PV guest
+       \item They will converge at some point, the agreed upon road map is to make HVMlite-nodm canonical "PVH"
+       \item End users probably won't notice the difference
+    \end{itemize}
+\end{frame}
+\note{
+    \begin{itemize}
+    \end{itemize}
+}
+
+\begin{frame}
+    \frametitle{Guest support}
+
+    \begin{itemize}
+        \item List of OSes and virtualization support:
+    \end{itemize}
+
+    {\scriptsize
+    \begin{table}
+    \centering
+    \tabulinesep=1.2mm
+    \begin{tabu} {| l | c | c | c | c | c |}
+    \cline{2-6}
+    \multicolumn{1}{c|}{}
+    & PV
+    & PVHVM
+    & HVM with PV drivers
+    & HVM
+    & PVH*
+    \\
+    \hline
+        Linux &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES \\
+    \hline
+        Windows &
+        \cellcolor{red!80}NO &
+        \cellcolor{red!80}NO &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{red!80}NO \\
+    \hline
+        NetBSD &
+        \cellcolor{green!80}YES &
+        \cellcolor{red!80}NO &
+        \cellcolor{red!80}NO &
+        \cellcolor{green!80}YES &
+        \cellcolor{red!80}NO \\
+    \hline
+        FreeBSD &
+        \cellcolor{red!80}NO &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{green!80}YES \\
+    \hline
+        OpenBSD &
+        \cellcolor{red!80}NO &
+        \cellcolor{yellow!80}YES &
+        \cellcolor{yellow!80}YES &
+        \cellcolor{green!80}YES &
+        \cellcolor{red!80}NO \\
+    \hline
+        DragonflyBSD &
+        \cellcolor{red!80}NO &
+        \cellcolor{red!80}NO &
+        \cellcolor{red!80}NO &
+        \cellcolor{green!80}YES &
+        \cellcolor{red!80}NO \\
+    \hline
+    \end{tabu}
+    \end{table}}
+
+\end{frame}
+\note {
+    \begin{itemize}
+        \item Linux and NetBSD are the only OSes to support PV.
+        \item Windows performance can be improved by installing PV drivers for network and disk devices.
+    \end{itemize}
+}
+
+\begin{frame}
+    \frametitle{Better scalability}
+    \begin{itemize}
+        \item Finer grained locks in hypervisor: per-vcpu maptrack free lists, per-cpu rwlock
+        \item Fairer locks in hypervisor: queue rwlock
+       \item Should benefit all guests, especially Xen virtual devices with multiqueue support (net, block)
+       \begin{itemize}
+               \item 2-socket Haswell-EP systems, Linux 16 queues inter-VM network throughput jumped from 15Gb/s to 48Gb/s
+       \end{itemize}
+    \end{itemize}
+\end{frame}
+\note{
+    \begin{itemize}
+
+    eede22972fefa02100226252c430ffcca99025eb
+
+    On multi-socket systems, the contention results in the locked compare swap
+    operation failing frequently which results in a tight loop of retries of the
+    compare swap operation. As the coherency fabric can only support a specific
+    rate of compare swap operations for a particular data location then taking
+    the read lock itself becomes a bottleneck for grant operations.
+    
+    Standard rwlock performance of a single VIF VM-VM transfer with 16 queues
+    configured was limited to approximately 15 gbit/s on a 2 socket Haswell-EP
+    host.
+    
+    Percpu rwlock performance with the same configuration is approximately
+    48 gbit/s.
+
+
+    \end{itemize}
+}
+
+
+\begin{frame}
+    \frametitle{Virtual Performance Monitoring Unit}
+    \begin{itemize}
+        \item Fully implemented in Xen 4.6, works for both PV and HVM
+       \item Intended for non-production use
+        \item Use dtrace or pmcstat to profile your VM
+    \end{itemize}
+    \begin{center}
+        \includegraphics[width=0.66\textwidth]{netfront-flame-graph.pdf}
+    \end{center}
+\end{frame}
+\note{
+    \begin{itemize}
+       \item intended for non-production use, implication is that there is security-wise support from upstream
+    \end{itemize}
+}
+
+\begin{frame}
+    \frametitle{xSplice - hypervisor hot-patching}
+    \begin{itemize}
+        \item ABC
+    \end{itemize}
+\end{frame}
+\note{
+    \begin{itemize}
+    \end{itemize}
+}
+
+
 \section{Live demo}
 
 \subsection{live}