\usepackage{amssymb,amsmath,epsfig,alltt}
\sloppy
\usepackage{palatino}
+\usepackage{url}
\usepackage{pdftricks}
\begin{psinputs}
\usepackage{pstricks}
We present an alternative approach which translate MIPS binaries
produced by any compiler into safe virtual machine bytecodes. This
approach offers four key advantages over existing techniques: it is
-language agnostic, it offers bug-for-bug compiler compatability,
+language agnostic, it offers bug-for-bug compiler compatibility,
requires no post-translation human intervention, and introduces no
build process modifications.
\section{Introduction}
-Unsafe languages such as C \cite{KR} and C++ \cite{soustroup} have
-been in use much longer than any of today's widely accepted safe
-languages such as Java \cite{java} and C\# \cite{csharp}.
+Unsafe languages such as C and C++ have been in use much longer than
+any of today's widely accepted safe languages such as Java and C\#
Consequently, there is a huge library of software written in these
languages. Although safe languages offer substantial benefits, their
comparatively young age often puts them at a disadvantage when breadth
performing transformations between them. In the third section we
introduce NestedVM and cover its two primary translation modes in
detail. Section four describes the NestedVM runtime, which plays the
-role of the OS kernel. Section five adresses the optimizations we
+role of the OS kernel. Section five addresses the optimizations we
employ and quantifies NestedVM's performance. Section six reviews our
experiences in applying NestedVM to various popular software packages.
We conclude with an analysis of NestedVM's weaknesses and potential
\subsubsection{Human-Assisted Translation}
Jazillian \cite{jazillian} is a commercial solution which produces
-extremely readable Java source code from C source code, but ony
+extremely readable Java source code from C source code, but only
translates a small portion of the C language. Jazillian is unique in
that in addition to {\it language migration}, it also performs {\it
API migration}; for example, Jazillian is intelligent enough to
translate ``{\tt char*~s1~=~strcpy(s2)}'' into ``{\tt
-String~s1~=~s2}''. Unfortunately such deep analysis is intractible
+String~s1~=~s2}''. Unfortunately such deep analysis is intractable
for most of the C language and standard library; indeed, Jazillian's
documentation notes that {\it ``This is not your father's language
translator... Jazillian does not always produce code that works
\subsubsection{Partial-Domain Translation}
-The {\tt c2j} \cite{c2j}, {\tt c2j++} \cite{c2jpp}, Cappucinno
+The {\tt c2j} \cite{c2j}, {\tt c2j++}, Cappucinno
\cite{capp}, and Ephedra \cite{ephedra} systems each provide support
for complete translation of a {\it subset} of the source language (C
or C++). Each of the four tools supports a progressively greater
Unfortunately, when the program being translated is large and complex,
it is quite likely that it will use an unsupported feature in at least
one place. In the absence of a programmer who understands the source
-program, a single anomoly is often enough to render the entire
+program, a single anomaly is often enough to render the entire
translation process useless. As a result, these tools are mainly
useful as an {\it aid} to programmers who could normally perform the
conversion themselves, but want to save time by automating most of the
which cannot be directly supported; thus the compiler still fails for
a substantial class of input programs.
-A Java backend for the {\tt lcc} [CITE] compiler, known as {\tt
-lcc-java} [CITE], is also available. Although this system is quite
+A Java backend for the {\tt lcc} compiler \cite{lcc}, known as {\tt
+lcc-java}, is also available. Although this system is quite
clean and elegantly designed, it lacks any form of system library
({\tt libc}), so very few C programs will run without custom
modification (which would cause them to diverge from the upstream
code generation steps of compilation, it is freed from the
extremely complex task of faithfully implementing languages
which are often not fully or formally specified (such as C and
- C++), and is able to support any langage for which a
+ C++), and is able to support any language for which a
MIPS-targeted compiler exists.
-\item {\bf Bug-for-bug compiler compatability}
+\item {\bf Bug-for-bug compiler compatibility}
Since NestedVM uses the compiler's {\it output} as its own {\it
input}, it ensures that programs which are inadvertently
emits a {\tt .java} file.
\item The resulting {\tt .java} code is compiled into a {\tt .class}
- file using {\tt jikes} \cite{jikes} or {\tt javac}.
+ file using {\tt jikes} or {\tt javac}.
\item At runtime, the host Java code invokes the {\tt run()} method on
the generated class. This is equivalent to the {\tt main()}
\section{The NestedVM Runtime}
-The NestedVM runtime fills the role typically assmed by an OS Kernel.
+The NestedVM runtime fills the role typically assumed by an OS Kernel.
Communication between MIPS code and the runtime is mediated by the
{\tt SYSCALL} instruction, just as the {\tt libc}-kernel interface is
on other MIPS implementations.
-Two implemenations of the runtime are offered; a simple runtime with
+Two implementations of the runtime are offered; a simple runtime with
the minimum support required to comply with ANSI C, and a more
sophisticated runtime which emulates a large portion of the POSIX API.
The next performance improvement came from tuning the size of the
methods invoked from the trampoline. Trial and error led to the
-onclusion that HotSpot \cite{hotspot} -- the most widely deployed JVM
+conclusion that HotSpot \cite{hotspot} -- the most widely deployed JVM
-- performs best when 128 MIPS instructions are mapped to each method.
\epsfig{file=charts/chart1,width=3in}
\end{itemize}
After tuning method sizes, our next performance boost came from
-eliminating exraneous case branches, which yielded approximately a
+eliminating extraneous case branches, which yielded approximately a
10\%-25\% performance improvement. Having case statements before each
instruction prevents JIT compilers from being able to optimize across
instruction boundaries, since control flow can enter the body of a
{\tt N\_0x000100000} is a non-final field to prevent {\tt javac} from
inlining it). This was sufficient to get reasonably large binaries to
compile, and caused only a small (approximately 5\%) performance
-degredation and a similarly small increase in the size of the {\tt
+degradation and a similarly small increase in the size of the {\tt
.class} file. However, as we will see in the next section, compiling
directly to {\tt .class} files (without the intermediate {\tt .java}
file) eliminates this problem entirely.
\subsection{Overall Performance}
+All times are measured in seconds. All tests were performed on a dual
+1Ghz Macintosh G4 running Apple's latest JVM (Sun HotSpot JDK
+1.4.1). Each test was run 8 times within a single VM. The highest and
+lowest times were removed and the remaining 6 were averaged. In each
+case only the first run differed significantly from the rest.
+
+The {\tt libjpeg} test consisted of decoding a 1280x1024 jpeg and
+writing a tga. The {\tt mspack} test consisted of extracting all
+members from {\tt arial32.exe}, {\tt comic32.exe}, {\tt times32.exe},
+and {\tt verdan32.exe}. The {\tt libfreetype} test consisted of
+rendering ASCII characters 32-127 of {\tt Comic.TTF} at sizes from 8
+to 48 incrementing by 4 for a total of 950 glyphs.
+
\epsfig{file=charts/chart8,width=3in}
\epsfig{file=charts/chart7,width=3in}
equivalent exists. The first is the FreeType font library, which
parses, hints, and rasterizes TrueType and Postscript fonts with
exceptional quality. The project also needed an open source JPEG
-decompressor; surprisingly, none exist for Java. While encoders are
+decompresser; surprisingly, none exist for Java. While encoders are
plentiful, decoders are rare, since Sun's J2SE VM includes a native
method to invoke {\tt libjpeg}.
Although we have only implemented it for the Java Virtual Machine, our
technique generalizes to other safe bytecode architectures. In
-particular we would like to demonstrate this generality by retargeting
+particular we would like to demonstrate this generality by re-targeting
the translator to the Microsoft Intermediate Language \cite{msil}.
Additionally, we would like to explore other uses for dynamic loading
http://nestedvm.ibex.org
\end{verbatim}
-\appendix
-\section{Appendix: Testing Methodology}
-
-All times are measured in seconds. All tests were performed on a dual
-1Ghz Macintosh G4 running Apple's latest JVM (Sun HotSpot JDK
-1.4.1). Each test was run 8 times within a single VM. The highest and
-lowest times were removed and the remaining 6 were averaged. In each
-case only the first run differed significantly from the rest.
-
-The {\tt libjpeg} test consisted of decoding a 1280x1024 jpeg and
-writing a tga. The {\tt mspack} test consisted of extracting all
-members from {\tt arial32.exe}, {\tt comic32.exe}, {\tt times32.exe},
-and {\tt verdan32.exe}. The {\tt libfreetype} test consisted of
-rendering ASCII characters 32-127 of {\tt Comic.TTF} at sizes from 8
-to 48 incrementing by 4 for a total of 950 glyphs.
-\bibliography{nestedvm}
+\bibliography{ivme04}
\end{document}
projects / nestedvm.git / blobdiff