\documentclass{article}

\title{Virtual Memory Managment for Large Address Spaces}
\author{Ian Wienand}

\begin{document}
\maketitle

The overhead of maintaining virtual memory can be a significant
inhibitor to system performance.

One approach to removing overheads is to use increased page sizes to
reduce the costs associated with TLB misses.  Using larger sized
virtual memory pages than in the base system is referred to as a
``superpage'' and has many research implications.  We are particularly
interested in researching the performance considerations with Linux on
the Intel Itanium architecture.

It is not a priori evident that always allocating superpages will lead
to greater performance as it has a range of side effects such as
increased cache usage and memory fragmentation.  When allocating
superpages there are many areas to take into account; allocation,
splitting and merging techniques are all areas for research.

The best ways to implement superpages is also a key area for research.
There are several strategies for implementing superpages.  Existing
approaches are generally static, increasing the page size in a uniform
manner.  The effects of dynamic allocation (in response to system
memory allocation) needs to be quantified.  Much more novel approaches
such as a guarded page table offer more research opportunities.

A general focus of the research relates to scalability with the
Itanium architecture.  All research areas need to have a focus on
behaviour in non-uniform memory architecture (NUMA) systems as this is
a primary application for the processor.  Some work will need to be
done with Linux to lay foundations for the research, particularly with
relation to the long format virtual hashed page table implementation
which is critical to using superpages on the architecture.

\section{About Ian}

Ian has been with Gelato@UNSW for three years working on Linux Itanium
scalability issues.  Currently he is working on foundation work for
superpage implementations.

\end{document}