Ignore:
Timestamp:
Jul 12, 2013, 1:03:13 PM (6 years ago)
Author:
cameron
Message:

Various revisions to handle referee comments

File:
1 edited

Legend:

Unmodified
Added
Removed
  • docs/Balisage13/Bal2013came0601/Bal2013came0601.html

    r3394 r3395  
    33<head>
    44<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
    5 <title></title>
     5<title>icXML:  Accelerating a Commercial XML
     6     Parser Using SIMD and Multicore Technologies</title>
    67<link rel="stylesheet" href="balisage-proceedings.css" type="text/css">
    78<meta name="keywords" content="">
     
    1314<html lang="en">
    1415<head>
    15 <title></title>
     16<title>icXML:  Accelerating a Commercial XML
     17     Parser Using SIMD and Multicore Technologies</title>
    1618<link rel="stylesheet" href="balisage-proceedings.css" type="text/css">
    1719<meta name="generator" content="Balisage Conference Proceedings XSLT (v1.2)">
     
    240242</div>
    241243<div id="mast"><div class="content">
    242 <h2 class="article-title" id="idp77056"></h2>
     244<h2 class="article-title" id="idp76432">icXML:  Accelerating a Commercial XML
     245     Parser Using SIMD and Multicore Technologies</h2>
    243246<div class="author">
    244247<h3 class="author">Nigel Medforth</h3>
     
    289292<h5 class="author-email"><code class="email">&lt;<a class="email" href="mailto:ashriram.cs.sfu.ca">ashriram.cs.sfu.ca</a>&gt;</code></h5>
    290293</div>
     294<div class="legalnotice-block"><p id="idp284896">Copyright © 2013 Nigel Medforth, Dan Lin, Kenneth S. Herdy, Robert D. Cameron  and Arrvindh Shriraman.
     295            This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative
     296            Works 2.5 Canada License.</p></div>
    291297<div class="mast-box">
    292 <p class="title"><a href="javascript:toggle('idp77616')" class="quiet"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp77616"></a> <span onclick="javascript:toggle('idp77616');return true">Abstract</span></p>
    293 <div class="folder" id="folder-idp77616" style="display:none"><p id="idp77920">Prior research on the acceleration of XML processing using SIMD and multi-core
    294             parallelism has lead to a number of interesting research prototypes. This work
    295             investigates the extent to which the techniques underlying these prototypes could result
    296             in systematic performance benefits when fully integrated into a commercial XML parser.
     298<p class="title"><a href="javascript:toggle('idp77216')" class="quiet"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp77216"></a> <span onclick="javascript:toggle('idp77216');return true">Abstract</span></p>
     299<div class="folder" id="folder-idp77216" style="display:none"><p id="idp77520">Prior research on the acceleration of XML processing using single-instruction
     300           multiple-data (SIMD) and multi-core
     301            parallelism has lead to a number of interesting research prototypes. This work is
     302            the first to investigate to the extent to which the techniques underlying these prototypes
     303            could result
     304            in systematic performance benefits when fully integrated into a commercial XML parser
    297305            The widely used Xerces-C++ parser of the Apache Software Foundation was chosen as the
    298306            foundation for the study. A systematic restructuring of the parser was undertaken, while
     
    300308            an increase in parsing speed of at least 50% was observed in a range of applications.
    301309            When coupled with pipeline parallelism on dual core processors, improvements of 2x and
    302             beyond were realized. </p></div>
     310            beyond were realized.
     311           
     312            icXML is intended as an important industrial contribution in its own right as well
     313            as an important case study for the underlying Parabix parallel processing framework.
     314            Based on the success of the icXML development, there is a strong case for continued
     315            development of that framework as well as for the application of that framework
     316            to other important XML technology stacks.   An important area for further work is
     317            the extension of Parabix technology to accelerate Java-based implementations as
     318            well as ones based on C/C++.
     319           
     320            </p></div>
    303321</div>
    304322<div class="toc">
    305323<p><b>Table of Contents</b></p>
    306324<dl>
    307 <dt><span class="section"><a href="#idp285392" class="toc">Introduction</a></span></dt>
     325<dt><span class="section"><a href="#idp286832" class="toc">Introduction</a></span></dt>
    308326<dt><span class="section"><a href="#background" class="toc">Background</a></span></dt>
    309327<dd><dl>
    310328<dt><span class="section"><a href="#background-xerces" class="toc">Xerces C++ Structure</a></span></dt>
    311 <dt><span class="section"><a href="#idp359776" class="toc">The Parabix Framework</a></span></dt>
    312 <dt><span class="section"><a href="#idp454528" class="toc">Sequential vs. Parallel Paradigm</a></span></dt>
     329<dt><span class="section"><a href="#idp361744" class="toc">The Parabix Framework</a></span></dt>
     330<dt><span class="section"><a href="#idp457376" class="toc">Sequential vs. Parallel Paradigm</a></span></dt>
    313331</dl></dd>
    314332<dt><span class="section"><a href="#architecture" class="toc">Architecture</a></span></dt>
    315333<dd><dl>
    316 <dt><span class="section"><a href="#idp462208" class="toc">Overview</a></span></dt>
     334<dt><span class="section"><a href="#idp465008" class="toc">Overview</a></span></dt>
    317335<dt><span class="section"><a href="#character-set-adapter" class="toc">Character Set Adapters</a></span></dt>
    318336<dt><span class="section"><a href="#par-filter" class="toc">Combined Parallel Filtering</a></span></dt>
     
    324342<dt><span class="section"><a href="#performance" class="toc">Performance</a></span></dt>
    325343<dd><dl>
    326 <dt><span class="section"><a href="#idp625760" class="toc">Xerces C++ SAXCount</a></span></dt>
    327 <dt><span class="section"><a href="#idp652080" class="toc">GML2SVG</a></span></dt>
     344<dt><span class="section"><a href="#idp654464" class="toc">Xerces C++ SAXCount</a></span></dt>
     345<dt><span class="section"><a href="#idp680992" class="toc">GML2SVG</a></span></dt>
    328346</dl></dd>
    329347<dt><span class="section"><a href="#conclusion" class="toc">Conclusion and Future Work</a></span></dt>
     
    331349</div>
    332350<div class="mast-box">
    333 <p class="title"><a href="javascript:toggle('idp79344')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp79344"></a> <span onclick="javascript:toggle('idp79344');return true">Nigel Medforth</span></p>
    334 <div class="folder" id="folder-idp79344" style="display:none">
     351<p class="title"><a href="javascript:toggle('idp79584')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp79584"></a> <span onclick="javascript:toggle('idp79584');return true">Nigel Medforth</span></p>
     352<div class="folder" id="folder-idp79584" style="display:none">
    335353<h5 class="author-email"><code class="email">&lt;<a class="email" href="mailto:nmedfort@sfu.ca">nmedfort@sfu.ca</a>&gt;</code></h5>
    336354<div class="affiliation">
     
    343361</div>
    344362<div class="personblurb">
    345 <p id="idp61664">Nigel Medforth is a M.Sc. student at Simon Fraser University and the lead
     363<p id="idp61840">Nigel Medforth is a M.Sc. student at Simon Fraser University and the lead
    346364               developer of icXML. He earned a Bachelor of Technology in Information Technology at
    347365               Kwantlen Polytechnic University in 2009 and was awarded the Dean’s Medal for
    348366               Outstanding Achievement.</p>
    349 <p id="idp62672">Nigel is currently researching ways to leverage both the Parabix framework and
     367<p id="idp62848">Nigel is currently researching ways to leverage both the Parabix framework and
    350368               stream-processing models to further accelerate XML parsing within icXML.</p>
    351369</div>
     
    353371</div>
    354372<div class="mast-box">
    355 <p class="title"><a href="javascript:toggle('idp66272')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp66272"></a> <span onclick="javascript:toggle('idp66272');return true">Dan Lin</span></p>
    356 <div class="folder" id="folder-idp66272" style="display:none">
     373<p class="title"><a href="javascript:toggle('idp66496')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp66496"></a> <span onclick="javascript:toggle('idp66496');return true">Dan Lin</span></p>
     374<div class="folder" id="folder-idp66496" style="display:none">
    357375<h5 class="author-email"><code class="email">&lt;<a class="email" href="mailto:lindanl@sfu.ca">lindanl@sfu.ca</a>&gt;</code></h5>
    358376<div class="affiliation">
     
    360378<p class="orgname">School of Computing Science, Simon Fraser University </p>
    361379</div>
    362 <div class="personblurb"><p id="idp67984">Dan Lin is a Ph.D student at Simon Fraser University. She earned a Master of Science
     380<div class="personblurb"><p id="idp68208">Dan Lin is a Ph.D student at Simon Fraser University. She earned a Master of Science
    363381             in Computing Science at Simon Fraser University in 2010. Her research focus on on high
    364382             performance algorithms that exploit parallelization strategies on various multicore platforms.
     
    367385</div>
    368386<div class="mast-box">
    369 <p class="title"><a href="javascript:toggle('idp70528')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp70528"></a> <span onclick="javascript:toggle('idp70528');return true">Kenneth Herdy</span></p>
    370 <div class="folder" id="folder-idp70528" style="display:none">
     387<p class="title"><a href="javascript:toggle('idp70752')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp70752"></a> <span onclick="javascript:toggle('idp70752');return true">Kenneth Herdy</span></p>
     388<div class="folder" id="folder-idp70752" style="display:none">
    371389<h5 class="author-email"><code class="email">&lt;<a class="email" href="mailto:ksherdy@sfu.ca">ksherdy@sfu.ca</a>&gt;</code></h5>
    372390<div class="affiliation">
     
    375393</div>
    376394<div class="personblurb">
    377 <p id="idp72256"> Ken Herdy completed an Advanced Diploma of Technology in Geographical Information
     395<p id="idp271952"> Ken Herdy completed an Advanced Diploma of Technology in Geographical Information
    378396               Systems at the British Columbia Institute of Technology in 2003 and earned a Bachelor
    379397               of Science in Computing Science with a Certificate in Spatial Information Systems at
    380398               Simon Fraser University in 2005. </p>
    381 <p id="idp72640"> Ken is currently pursuing PhD studies in Computing Science at Simon Fraser
     399<p id="idp272688"> Ken is currently pursuing PhD studies in Computing Science at Simon Fraser
    382400               University with industrial scholarship support from the Natural Sciences and
    383401               Engineering Research Council of Canada, the Mathematics of Information Technology and
     
    389407</div>
    390408<div class="mast-box">
    391 <p class="title"><a href="javascript:toggle('idp274496')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp274496"></a> <span onclick="javascript:toggle('idp274496');return true">Rob Cameron</span></p>
    392 <div class="folder" id="folder-idp274496" style="display:none">
     409<p class="title"><a href="javascript:toggle('idp275424')" class="linkbox"><img class="toc-icon" src="plus.png" alt="expand" id="icon-idp275424"></a> <span onclick="javascript:toggle('idp275424');return true">Rob Cameron</span></p>
     410<div class="folder" id="folder-idp275424" style="display:none">
    393411<h5 class="author-email"><code class="email">&lt;<a class="email" href="mailto:cameron@cs.sfu.ca">cameron@cs.sfu.ca</a>&gt;</code></h5>
    394412<div class="affiliation">
     
    400418<p class="orgname">International Characters, Inc.</p>
    401419</div>
    402 <div class="personblurb"><p id="idp276160">Dr. Rob Cameron is Professor of Computing Science and Associate Dean of Applied
     420<div class="personblurb"><p id="idp277088">Dr. Rob Cameron is Professor of Computing Science and Associate Dean of Applied
    403421               Sciences at Simon Fraser University. His research interests include programming
    404422               language and software system technology, with a specific focus on high performance
     
    416434<div id="main">
    417435<div class="article">
    418 <h2 class="article-title" id="idp77056"></h2>
    419 <div class="section" id="idp285392">
     436<h2 class="article-title" id="idp76432">icXML:  Accelerating a Commercial XML
     437     Parser Using SIMD and Multicore Technologies</h2>
     438<div class="section" id="idp286832">
    420439<h2 class="title" style="clear: both">Introduction</h2>
    421 <p id="idp286032">   
     440<p id="idp287472">   
    422441        Parallelization and acceleration of XML parsing is a widely
    423442        studied problem that has seen the development of a number
     
    429448        For example, one possibility for data
    430449        parallelization is to add a pre-parsing step to compute
    431         a skeleton tree structure of an  XML document <a class="xref" id="idp286848" href="javascript:showcite('cite-GRID2006','idp286848')">Lu and Chiu 2006</a>.
     450        a skeleton tree structure of an  XML document <a class="xref" id="idp288288" href="javascript:showcite('cite-GRID2006','idp288288')">[Lu and Chiu 2006]</a>.
    432451        The parallelization of the pre-parsing stage itself can be tackled with
    433           state machines <a class="xref" id="idp299760" href="javascript:showcite('cite-E-SCIENCE2007','idp299760')">Pan and Zhang 2007</a>, <a class="xref" id="idp300352" href="javascript:showcite('cite-IPDPS2008','idp300352')">Pan and Zhang 2008</a>.
    434         Methods without pre-parsing have used speculation <a class="xref" id="idp301200" href="javascript:showcite('cite-HPCC2011','idp301200')">You and Wang 2011</a> or post-processing that
    435         combines the partial results <a class="xref" id="idp302032" href="javascript:showcite('cite-ParaDOM2009','idp302032')">Shah and Rao 2009</a>.
     452          state machines <a class="xref" id="idp301312" href="javascript:showcite('cite-E-SCIENCE2007','idp301312')">[Pan and Zhang 2007]</a>, <a class="xref" id="idp302064" href="javascript:showcite('cite-IPDPS2008','idp302064')">[Pan and Zhang 2008b]</a>.
     453        Methods without pre-parsing have used speculation <a class="xref" id="idp302880" href="javascript:showcite('cite-HPCC2011','idp302880')">[You and Wang 2011]</a> or post-processing that
     454        combines the partial results <a class="xref" id="idp303712" href="javascript:showcite('cite-ParaDOM2009','idp303712')">[Shah and Rao 2009]</a>.
    436455        A hybrid technique that combines data and pipeline parallelism was proposed to
    437         hide the latency of a "job" that has to be done sequentially <a class="xref" id="idp302896" href="javascript:showcite('cite-ICWS2008','idp302896')">Pan and Zhang 2008</a>.
     456        hide the latency of a "job" that has to be done sequentially <a class="xref" id="idp304576" href="javascript:showcite('cite-ICWS2008','idp304576')">[Pan and Zhang 2008a]</a>.
    438457      </p>
    439 <p id="idp303776">
     458<p id="idp305456">
    440459        Fewer efforts have investigated SIMD parallelism, although this approach
    441460        has the potential advantage of improving single core performance as well
    442         as offering savings in energy consumption <a class="xref" id="idp304240" href="javascript:showcite('cite-HPCA2012','idp304240')">Lin and Medforth 2012</a>.
     461        as offering savings in energy consumption <a class="xref" id="idp305920" href="javascript:showcite('cite-HPCA2012','idp305920')">[Lin and Medforth 2012]</a>.
    443462        Intel introduced specialized SIMD string processing instructions in the SSE 4.2 instruction set extension
    444         and showed how they can be used to improve the performance of XML parsing <a class="xref" id="idp305184" href="javascript:showcite('cite-XMLSSE42','idp305184')">Lei 2008</a>.
     463        and showed how they can be used to improve the performance of XML parsing <a class="xref" id="idp306864" href="javascript:showcite('cite-XMLSSE42','idp306864')">[Lei 2008]</a>.
    445464        The Parabix framework uses generic SIMD extensions and bit parallel methods to
    446         process hundreds of XML input characters simultaneously <a class="xref" id="idp306096" href="javascript:showcite('cite-Cameron2009','idp306096')">Cameron and Herdy 2009</a> <a class="xref" id="idp306848" href="javascript:showcite('cite-cameron-EuroPar2011','idp306848')">Cameron and Amiri 2011</a>.
     465        process hundreds of XML input characters simultaneously <a class="xref" id="idp307776" href="javascript:showcite('cite-Cameron2009','idp307776')">[Balisage 2009]</a> <a class="xref" id="idp308528" href="javascript:showcite('cite-cameron-EuroPar2011','idp308528')">[Parabix2 2011]</a>.
    447466        Parabix prototypes have also combined SIMD methods with thread-level parallelism to
    448         achieve further acceleration on multicore systems <a class="xref" id="idp307760" href="javascript:showcite('cite-HPCA2012','idp307760')">Lin and Medforth 2012</a>.
     467        achieve further acceleration on multicore systems <a class="xref" id="idp309440" href="javascript:showcite('cite-HPCA2012','idp309440')">[Lin and Medforth 2012]</a>.
    449468      </p>
    450 <p id="idp308528">
     469<p id="idp310208">
    451470        In this paper, we move beyond research prototypes to consider
    452471        the detailed integration of both SIMD and multicore parallelism into the
     
    467486        multiple cores.
    468487      </p>
    469 <p id="idp309968">
     488<p id="idp311648">
    470489        The remainder of this paper is organized as follows.   
    471490          <a class="xref" href="#background" title="Background">section “Background”</a> discusses the structure of the Xerces and Parabix XML parsers and the fundamental
     
    487506<div class="section" id="background-xerces">
    488507<h3 class="title" style="clear: both">Xerces C++ Structure</h3>
    489 <p id="idp317232"> The Xerces C++ parser is a widely-used standards-conformant
     508<p id="idp318976"> The Xerces C++ parser is a widely-used standards-conformant
    490509            XML parser produced as open-source software
    491510             by the Apache Software Foundation.
     
    498517            parsing using either pull parsing or SAX/SAX2 push-style parsing as well as a DOM
    499518            tree-based parsing interface. </p>
    500 <p id="idp318496">
     519<p id="idp321104">
    501520            Xerces,
    502521            like all traditional parsers, processes XML documents sequentially a byte-at-a-time from
     
    517536<div class="table-wrapper" id="xerces-profile">
    518537<p class="title">Table I</p>
    519 <div class="caption"><p id="idp13600">Execution Time of Top 10 Xerces Functions</p></div>
     538<div class="caption"><p id="idp13392">Execution Time of Top 10 Xerces Functions</p></div>
    520539<table class="table" xml:id="xerces-profile">
    521540<colgroup span="1">
     
    572591</div>
    573592</div>
    574 <div class="section" id="idp359776">
     593<div class="section" id="idp361744">
    575594<h3 class="title" style="clear: both">The Parabix Framework</h3>
    576 <p id="idp360416"> The Parabix (parallel bit stream) framework is a transformative approach to XML
     595<p id="idp362384"> The Parabix (parallel bit stream) framework is a transformative approach to XML
    577596            parsing (and other forms of text processing.) The key idea is to exploit the
    578597            availability of wide SIMD registers (e.g., 128-bit) in commodity processors to represent
     
    603622<div class="table-wrapper" id="xml-bytes">
    604623<p class="title">Table II</p>
    605 <div class="caption"><p id="idp374112">XML Source Data</p></div>
     624<div class="caption"><p id="idp376256">XML Source Data</p></div>
    606625<table class="table" xml:id="xml-bytes">
    607626<colgroup span="1">
     
    632651<div class="table-wrapper" id="xml-bits">
    633652<p class="title">Table III</p>
    634 <div class="caption"><p id="idp390384">8-bit ASCII Basis Bit Streams</p></div>
     653<div class="caption"><p id="idp392528">8-bit ASCII Basis Bit Streams</p></div>
    635654<table class="table" xml:id="xml-bits">
    636655<colgroup span="1">
     
    699718</table>
    700719</div>
    701 <p id="idp430512"> Consider, for example, the XML source data stream shown in the first line of <a class="xref" href="#derived">Table IV</a>.
     720<p id="idp432656"> Consider, for example, the XML source data stream shown in the first line of <a class="xref" href="#derived">Table IV</a>.
    702721The remaining lines of this figure show
    703722            several parallel bit streams that are computed in Parabix-style parsing, with each bit
     
    709728            character immediately following the opener (i.e., "<code class="code">/</code>") or
    710729            not. The remaining three lines show streams that can be computed in subsequent parsing
    711             (using the technique of bitstream addition <a class="xref" id="idp433616" href="javascript:showcite('cite-cameron-EuroPar2011','idp433616')">Cameron and Amiri 2011</a>), namely streams
     730            (using the technique of bitstream addition <a class="xref" id="idp435760" href="javascript:showcite('cite-cameron-EuroPar2011','idp435760')">[Parabix2 2011]</a>), namely streams
    712731            marking the element names, attribute names and attribute values of tags. </p>
    713732<div class="table-wrapper" id="derived">
    714733<p class="title">Table IV</p>
    715 <div class="caption"><p id="idp435328">XML Source Data and Derived Parallel Bit Streams</p></div>
     734<div class="caption"><p id="idp437472">XML Source Data and Derived Parallel Bit Streams</p></div>
    716735<table class="table" xml:id="derived">
    717736<colgroup span="1">
     
    763782</table>
    764783</div>
    765 <p id="idp448432"> Two intuitions may help explain how the Parabix approach can lead to improved XML
     784<p id="idp450576"> Two intuitions may help explain how the Parabix approach can lead to improved XML
    766785            parsing performance. The first is that the use of the full register width offers a
    767786            considerable information advantage over sequential byte-at-a-time parsing. That is,
     
    772791            individual decision-bits, an approach that computes many of them in parallel (e.g., 128
    773792            bytes at a time using 128-bit registers) should provide substantial benefit. </p>
    774 <p id="idp449680"> Previous studies have shown that the Parabix approach improves many aspects of XML
    775             processing, including transcoding <a class="xref" id="idp450080" href="javascript:showcite('cite-Cameron2008','idp450080')">Cameron 2008</a>, character classification and
     793<p id="idp452688"> Previous studies have shown that the Parabix approach improves many aspects of XML
     794            processing, including transcoding <a class="xref" id="idp453088" href="javascript:showcite('cite-Cameron2008','idp453088')">[u8u16 2008]</a>, character classification and
    776795            validation, tag parsing and well-formedness checking. The first Parabix parser used
    777796            processor bit scan instructions to considerably accelerate sequential scanning loops for
    778             individual characters <a class="xref" id="idp451088" href="javascript:showcite('cite-CameronHerdyLin2008','idp451088')">Cameron and Herdy 2008</a>. Recent work has incorporated a method
    779             of parallel scanning using bitstream addition <a class="xref" id="idp451952" href="javascript:showcite('cite-cameron-EuroPar2011','idp451952')">Cameron and Amiri 2011</a>, as well as
     797            individual characters <a class="xref" id="idp453984" href="javascript:showcite('cite-CameronHerdyLin2008','idp453984')">[Parabix1 2008]</a>. Recent work has incorporated a method
     798            of parallel scanning using bitstream addition <a class="xref" id="idp454800" href="javascript:showcite('cite-cameron-EuroPar2011','idp454800')">[Parabix2 2011]</a>, as well as
    780799            combining SIMD methods with 4-stage pipeline parallelism to further improve throughput
    781             <a class="xref" id="idp452736" href="javascript:showcite('cite-HPCA2012','idp452736')">Lin and Medforth 2012</a>. Although these research prototypes handled the full syntax of
     800            <a class="xref" id="idp455584" href="javascript:showcite('cite-HPCA2012','idp455584')">[Lin and Medforth 2012]</a>. Although these research prototypes handled the full syntax of
    782801            schema-less XML documents, they lacked the functionality required by full XML parsers. </p>
    783 <p id="idp453680"> Commercial XML processors support transcoding of multiple character sets and can
     802<p id="idp456528"> Commercial XML processors support transcoding of multiple character sets and can
    784803            parse and validate against multiple document vocabularies. Additionally, they provide
    785804            API facilities beyond those found in research prototypes, including the widely used SAX,
    786805            SAX2 and DOM interfaces. </p>
    787806</div>
    788 <div class="section" id="idp454528">
     807<div class="section" id="idp457376">
    789808<h3 class="title" style="clear: both">Sequential vs. Parallel Paradigm</h3>
    790 <p id="idp455216"> Xerces—like all traditional XML parsers—processes XML documents
     809<p id="idp458064"> Xerces—like all traditional XML parsers—processes XML documents
    791810            sequentially. Each character is examined to distinguish between the XML-specific markup,
    792811            such as a left angle bracket <code class="code">&lt;</code>, and the content held within the
    793812            document. As the parser progresses through the document, it alternates between markup
    794813            scanning, validation and content processing modes. </p>
    795 <p id="idp456752"> In other words, Xerces belongs to an equivalence class of applications termed FSM
     814<p id="idp459600"> In other words, Xerces belongs to an equivalence class of applications termed FSM
    796815           applications<sup class="fn-label"><a href="#FSM" class="footnoteref">[1]</a></sup>.<sup class="fn-label"><a href="#FSM" class="footnoteref" id="FSM-ref">[1]</a></sup> Each state transition indicates the processing context of
    797816            subsequent characters. Unfortunately, textual data tends to be unpredictable and any
    798817            character could induce a state transition. </p>
    799 <p id="idp459280"> Parabix-style XML parsers utilize a concept of layered processing. A block of source
     818<p id="idp462080"> Parabix-style XML parsers utilize a concept of layered processing. A block of source
    800819            text is transformed into a set of lexical bitstreams, which undergo a series of
    801820            operations that can be grouped into logical layers, e.g., transposition, character
    802821            classification, and lexical analysis. Each layer is pipeline parallel and require
    803             neither speculation nor pre-parsing stages<a class="xref" id="idp459968" href="javascript:showcite('cite-HPCA2012','idp459968')">Lin and Medforth 2012</a>. To meet the API requirements
     822            neither speculation nor pre-parsing stages<a class="xref" id="idp462768" href="javascript:showcite('cite-HPCA2012','idp462768')">[Lin and Medforth 2012]</a>. To meet the API requirements
    804823            of the document-ordered Xerces output, the results of the Parabix processing layers must
    805824            be interleaved to produce the equivalent behaviour. </p>
     
    808827<div class="section" id="architecture">
    809828<h2 class="title" style="clear: both">Architecture</h2>
    810 <div class="section" id="idp462208">
     829<div class="section" id="idp465008">
    811830<h3 class="title" style="clear: both">Overview</h3>
    812 <p id="idp463264"> icXML is more than an optimized version of Xerces. Many components were grouped,
     831<p id="idp466064"> icXML is more than an optimized version of Xerces. Many components were grouped,
    813832            restructured and rearchitected with pipeline parallelism in mind. In this section, we
    814833            highlight the core differences between the two systems. As shown in Figure
     
    836855<p class="title">Figure 1: Xerces Architecture</p>
    837856<div class="figure-contents">
    838 <div class="mediaobject" id="idp471232"><img alt="png image (xerces.png)" src="xerces.png" width="150cm"></div>
     857<div class="mediaobject" id="idp474032"><img alt="png image (xerces.png)" src="xerces.png" width="150cm"></div>
    839858<div class="caption"></div>
    840859</div>
    841860</div>
    842 <p id="idp473552"> In icXML functions are grouped into logical components. As shown in
     861<p id="idp476352"> In icXML functions are grouped into logical components. As shown in
    843862             <a class="xref" href="#xerces-arch" title="Xerces Architecture">Figure 1</a>, two major categories exist: (1) the Parabix Subsystem and (2) the
    844                Markup Processor. All tasks in (1) use the Parabix Framework <a class="xref" id="idp474640" href="javascript:showcite('cite-HPCA2012','idp474640')">Lin and Medforth 2012</a>, which
     863               Markup Processor. All tasks in (1) use the Parabix Framework <a class="xref" id="idp477440" href="javascript:showcite('cite-HPCA2012','idp477440')">[Lin and Medforth 2012]</a>, which
    845864            represents data as a set of parallel bitstreams. The <span class="ital">Character Set
    846865              Adapter</span>, discussed in <a class="xref" href="#character-set-adapter" title="Character Set Adapters">section “Character Set Adapters”</a>, mirrors
    847866            Xerces's Transcoder duties; however instead of producing UTF-16 it produces a set of
    848               lexical bitstreams, similar to those shown in . These lexical
     867              lexical bitstreams, similar to those shown in <a class="xref" id="idp479904" href="javascript:showcite('cite-CameronHerdyLin2008','idp479904')">[Parabix1 2008]</a>. These lexical
    849868            bitstreams are later transformed into UTF-16 in the Content Stream Generator, after
    850869            additional processing is performed. The first precursor to producing UTF-16 is the
     
    859878              described in <a class="xref" href="#errorhandling" title="Error Handling">section “Error Handling”</a>. From here, two data-independent
    860879            branches exist: the Symbol Resolver and Content Preparation Unit. </p>
    861 <p id="idp481088"> A typical XML file contains few unique element and attribute names—but
     880<p id="idp483888"> A typical XML file contains few unique element and attribute names—but
    862881            each of them will occur frequently. icXML stores these as distinct data structures,
    863882            called symbols, each with their own global identifier (GID). Using the symbol marker
     
    865884               Resolver</span> scans through the raw data to produce a sequence of GIDs, called
    866885            the <span class="ital">symbol stream</span>. </p>
    867 <p id="idp483744"> The final components of the Parabix Subsystem are the <span class="ital">Content
     886<p id="idp486544"> The final components of the Parabix Subsystem are the <span class="ital">Content
    868887               Preparation Unit</span> and <span class="ital">Content Stream
    869888            Generator</span>. The former takes the (transposed) basis bitstreams and selectively
    870889            filters them, according to the information provided by the Parallel Markup Parser, and
    871890            the latter transforms the filtered streams into the tagged UTF-16 <span class="ital">content stream</span>, discussed in <a class="xref" href="#contentstream" title="Content Stream">section “Content Stream”</a>. </p>
    872 <p id="idp487344"> Combined, the symbol and content stream form icXML's compressed IR of the XML
    873             document. The <span class="ital">Markup Processor</span>~parses the IR to
     891<p id="idp490144"> Combined, the symbol and content stream form icXML's compressed IR of the XML
     892            document. The <span class="ital">Markup Processor</span>
     893            parses the IR to
    874894            validate and produce the sequential output for the end user. The <span class="ital">Final WF checker</span> performs inter-element well-formedness validation that
    875895            would be too costly to perform in bit space, such as ensuring every start tag has a
     
    882902<p class="title">Figure 2: icXML Architecture</p>
    883903<div class="figure-contents">
    884 <div class="mediaobject" id="idp493808"><img alt="png image (icxml.png)" src="icxml.png" width="500cm"></div>
     904<div class="mediaobject" id="idp496608"><img alt="png image (icxml.png)" src="icxml.png" width="500cm"></div>
    885905<div class="caption"></div>
    886906</div>
     
    889909<div class="section" id="character-set-adapter">
    890910<h3 class="title" style="clear: both">Character Set Adapters</h3>
    891 <p id="idp497296"> In Xerces, all input is transcoded into UTF-16 to simplify the parsing costs of
     911<p id="idp500096"> In Xerces, all input is transcoded into UTF-16 to simplify the parsing costs of
    892912            Xerces itself and provide the end-consumer with a single encoding format. In the
    893913            important case of UTF-8 to UTF-16 transcoding, the transcoding costs can be significant,
     
    896916            other cases, transcoding may involve table look-up operations for each byte of input. In
    897917            any case, transcoding imposes at least a cost of buffer copying. </p>
    898 <p id="idp498352"> In icXML, however, the concept of Character Set Adapters (CSAs) is used to minimize
     918<p id="idp501152"> In icXML, however, the concept of Character Set Adapters (CSAs) is used to minimize
    899919            transcoding costs. Given a specified input encoding, a CSA is responsible for checking
    900920            that input code units represent valid characters, mapping the characters of the encoding
     
    902922            item streams), as well as supporting ultimate transcoding requirements. All of this work
    903923            is performed using the parallel bitstream representation of the source input. </p>
    904 <p id="idp499328"> An important observation is that many character sets are an extension to the legacy
     924<p id="idp40944"> An important observation is that many character sets are an extension to the legacy
    905925            7-bit ASCII character set. This includes the various ISO Latin character sets, UTF-8,
    906926            UTF-16 and many others. Furthermore, all significant characters for parsing XML are
    907927            confined to the ASCII repertoire. Thus, a single common set of lexical item calculations
    908928            serves to compute lexical item streams for all such ASCII-based character sets. </p>
    909 <p id="idp500208"> A second observation is that—regardless of which character set is
     929<p id="idp41824"> A second observation is that—regardless of which character set is
    910930            used—quite often all of the characters in a particular block of input will be
    911931            within the ASCII range. This is a very simple test to perform using the bitstream
     
    914934            be skipped. Transcoding to UTF-16 becomes trivial as the high eight bitstreams of the
    915935            UTF-16 form are each set to zero in this case. </p>
    916 <p id="idp501648"> A third observation is that repeated transcoding of the names of XML elements,
     936<p id="idp43744"> A third observation is that repeated transcoding of the names of XML elements,
    917937            attributes and so on can be avoided by using a look-up mechanism. That is, the first
    918938            occurrence of each symbol is stored in a look-up table mapping the input encoding to a
     
    921941            symbol look up is required to apply various XML validation rules, there is achieves the
    922942            effect of transcoding each occurrence without additional cost. </p>
    923 <p id="idp502704"> The cost of individual character transcoding is avoided whenever a block of input is
     943<p id="idp44800"> The cost of individual character transcoding is avoided whenever a block of input is
    924944            confined to the ASCII subset and for all but the first occurrence of any XML element or
    925945            attribute name. Furthermore, when transcoding is required, the parallel bitstream
    926946            representation supports efficient transcoding operations. In the important case of UTF-8
    927947            to UTF-16 transcoding, the corresponding UTF-16 bitstreams can be calculated in bit
    928               parallel fashion based on UTF-8 streams <a class="xref" id="idp503504" href="javascript:showcite('cite-Cameron2008','idp503504')">Cameron 2008</a>, and all but the final bytes
     948              parallel fashion based on UTF-8 streams <a class="xref" id="idp45600" href="javascript:showcite('cite-Cameron2008','idp45600')">[u8u16 2008]</a>, and all but the final bytes
    929949            of multi-byte sequences can be marked for deletion as discussed in the following
    930950            subsection. In other cases, transcoding within a block only need be applied for
     
    934954<div class="section" id="par-filter">
    935955<h3 class="title" style="clear: both">Combined Parallel Filtering</h3>
    936 <p id="idp505856"> As just mentioned, UTF-8 to UTF-16 transcoding involves marking all but the last
     956<p id="idp47952"> As just mentioned, UTF-8 to UTF-16 transcoding involves marking all but the last
    937957            bytes of multi-byte UTF-8 sequences as positions for deletion. For example, the two
    938958            Chinese characters <code class="code">䜠奜</code> are represented as two
     
    947967            input bytes is the bit sequence <code class="code">110110</code>. Using this approach, transcoding
    948968            may then be completed by applying parallel deletion and inverse transposition of the
    949             UTF-16 bitstreams<a class="xref" id="idp509840" href="javascript:showcite('cite-Cameron2008','idp509840')">Cameron 2008</a>. </p>
    950 <p id="idp510640"> Rather than immediately paying the costs of deletion and transposition just for
     969            UTF-16 bitstreams<a class="xref" id="idp521312" href="javascript:showcite('cite-Cameron2008','idp521312')">[u8u16 2008]</a>. </p>
     970<p id="idp522112"> Rather than immediately paying the costs of deletion and transposition just for
    951971            transcoding, however, icXML defers these steps so that the deletion masks for several
    952972            stages of processing may be combined. In particular, this includes core XML requirements
     
    963983<div class="figure-contents">
    964984<div class="caption">Line Break Normalization Logic</div>
    965 <pre class="programlisting" id="idp514592">
     985<pre class="programlisting" id="idp525120">
    966986# XML 1.0 line-break normalization rules.
    967987if lex.CR:
     
    979999</div>
    9801000         </p>
    981 <p id="idp515936"> In essence, the deletion masks for transcoding and for line break normalization each
     1001<p id="idp526592"> In essence, the deletion masks for transcoding and for line break normalization each
    9821002            represent a bitwise filter; these filters can be combined using bitwise-or so that the
    9831003            parallel deletion algorithm need only be applied once. </p>
    984 <p id="idp516592"> A further application of combined filtering is the processing of XML character and
    985             entity references. Consider, for example, the references <code class="code">&amp;</code> or
    986                <code class="code">&lt;</code>. which must be replaced in XML processing with the single
     1004<p id="idp527248"> A further application of combined filtering is the processing of XML character and
     1005           entity references. Consider, for example, the references <code class="code">&amp;amp;</code> or
     1006             <code class="code">&amp;#x3C;</code> which must be replaced in XML processing with the single
    9871007               <code class="code">&amp;</code> and <code class="code">&lt;</code> characters, respectively. The
    9881008            approach in icXML is to mark all but the first character positions of each reference for
    9891009            deletion, leaving a single character position unmodified. Thus, for the references
    990                <code class="code">&amp;</code> or <code class="code">&lt;</code> the masks <code class="code">01111</code> and
     1010               <code class="code">&amp;amp;</code> or <code class="code">&amp;#x3C;</code> the masks <code class="code">01111</code> and
    9911011               <code class="code">011111</code> are formed and combined into the overall deletion mask. After the
    9921012            deletion and inverse transposition operations are finally applied, a post-processing
     
    9951015            UTF-16 code unit. In the case, that this is not true, it is addressed in
    9961016            post-processing. </p>
    997 <p id="idp521408"> The final step of combined filtering occurs during the process of reducing markup
     1017<p id="idp532096"> The final step of combined filtering occurs during the process of reducing markup
    9981018            data to tag bytes preceding each significant XML transition as described in
    9991019              <a class="xref" href="#contentstream" title="Content Stream">section “Content Stream”</a>. Overall, icXML avoids separate buffer copying
    10001020            operations for each of the these filtering steps, paying the cost of parallel deletion
    10011021            and inverse transposition only once. Currently, icXML employs the parallel-prefix
    1002             compress algorithm of Steele~<a class="xref" id="idp522832" href="javascript:showcite('cite-HackersDelight','idp522832')">Warren 2002</a> Performance is independent of the
     1022            compress algorithm of Steele <a class="xref" id="idp533408" href="javascript:showcite('cite-HackersDelight','idp533408')">[Warren 2002]</a>. Performance is independent of the
    10031023            number of positions deleted. Future versions of icXML are expected to take advantage of
    1004             the parallel extract operation~<a class="xref" id="idp523840" href="javascript:showcite('cite-HilewitzLee2006','idp523840')">Hilewitz and Lee 2006</a> that Intel is now providing in its
     1024            the parallel extract operation <a class="xref" id="idp534304" href="javascript:showcite('cite-HilewitzLee2006','idp534304')">[Hilewitz and Lee 2006]</a> that Intel is now providing in its
    10051025            Haswell architecture. </p>
    10061026</div>
    10071027<div class="section" id="contentstream">
    10081028<h3 class="title" style="clear: both">Content Stream</h3>
    1009 <p id="idp525888"> A relatively-unique concept for icXML is the use of a filtered content stream.
     1029<p id="idp536352"> A relatively-unique concept for icXML is the use of a filtered content stream.
    10101030            Rather that parsing an XML document in its original format, the input is transformed
    10111031            into one that is easier for the parser to iterate through and produce the sequential
    1012             output. In , the source data
    1013            
    1014             is transformed into
    1015            
     1032            output. In <a class="xref" href="#fig-parabix2">Table V</a>, the source data
     1033             <code class="code"> &lt;document&gt;fee&lt;element a1='fie' a2 = 'foe'&gt;&lt;/element&gt;fum&lt;/document&gt;</code>
     1034             is transformed into <code class="code"><span class="ital">0</span>fee<span class="ital">0</span>=fie<span class="ital">0</span>=foe<span class="ital">0</span>&gt;<span class="ital">0</span>/fum<span class="ital">0</span>/</code>
    10161035            through the parallel filtering algorithm, described in <a class="xref" href="#par-filter" title="Combined Parallel Filtering">section “Combined Parallel Filtering”</a>. </p>
    1017 <p id="idp529072"> Combined with the symbol stream, the parser traverses the content stream to
     1036<div class="table-wrapper" id="fig-parabix2">
     1037<p class="title">Table V</p>
     1038<div class="caption">XML Source Data and Derived Parallel Bit Streams</div>
     1039<table class="table" xml:id="fig-parabix2">
     1040<colgroup span="1">
     1041<col align="centre" valign="top" span="1">
     1042<col align="left" valign="top" span="1">
     1043</colgroup>
     1044<tbody>
     1045<tr>
     1046<td> Source Data </td>
     1047<td>
     1048                                    <code class="code"> &lt;document&gt;fee&lt;element a1='fie' a2 = 'foe'&gt;&lt;/element&gt;fum&lt;/document&gt; </code>
     1049</td>
     1050</tr>
     1051<tr>
     1052<td> String Ends </td>
     1053<td> <code class="code">1____________1_______________1__________1_1____________1__________</code>
     1054</td>
     1055</tr>
     1056<tr>
     1057<td> Markup Identifiers </td>
     1058<td>         <code class="code">_________1______________1_________1______1_1____________1_________</code>
     1059</td>
     1060</tr>
     1061<tr>
     1062<td> Deletion Mask </td>
     1063<td>              <code class="code">_11111111_____1111111111_1____1111_11_______11111111_____111111111</code>
     1064</td>
     1065</tr>
     1066<tr>
     1067<td> Undeleted Data </td>
     1068<td> <code class="code"><span class="ital">0</span>________&gt;fee<span class="ital">0</span>__________=_fie<span class="ital">0</span>____=__foe<span class="ital">0</span>&gt;<span class="ital">0</span>/________fum<span class="ital">0</span>/_________</code>
     1069</td>
     1070</tr>
     1071</tbody>
     1072</table>
     1073</div>
     1074<p id="idp557712"> Combined with the symbol stream, the parser traverses the content stream to
    10181075            effectively reconstructs the input document in its output form. The initial <span class="ital">0</span> indicates an empty content string. The following
    10191076               <code class="code">&gt;</code> indicates that a start tag without any attributes is the first
     
    10271084            null character in the content stream in parallel, which in turn means the parser can
    10281085            directly jump to the end of every string without scanning for it. </p>
    1029 <p id="idp533104"> Following <code class="code">'fee'</code> is a <code class="code">=</code>, which marks the
     1086<p id="idp561792"> Following <code class="code">'fee'</code> is a <code class="code">=</code>, which marks the
    10301087            existence of an attribute. Because all of the intra-element was performed in the Parabix
    10311088            Subsystem, this must be a legal attribute. Since attributes can only occur within start
     
    10431100<div class="section" id="namespace-handling">
    10441101<h3 class="title" style="clear: both">Namespace Handling</h3>
    1045 <p id="idp538672"> In XML, namespaces prevents naming conflicts when multiple vocabularies are used
     1102<p id="idp567360"> In XML, namespaces prevents naming conflicts when multiple vocabularies are used
    10461103            together. It is especially important when a vocabulary application-dependant meaning,
    10471104            such as when XML or SVG documents are embedded within XHTML files. Namespaces are bound
    10481105            to uniform resource identifiers (URIs), which are strings used to identify specific
    1049             names or resources. On line 1 in <a class="xref" href="#namespace-ex">Table V</a>, the <code class="code">xmlns</code>
     1106            names or resources. On line 1 in <a class="xref" href="#namespace-ex">Table VI</a>, the <code class="code">xmlns</code>
    10501107            attribute instructs the XML processor to bind the prefix <code class="code">p</code> to the URI
    10511108               '<code class="code">pub.net</code>' and the default (empty) prefix to
     
    10611118            that are in-scope. </p>
    10621119<div class="table-wrapper" id="namespace-ex">
    1063 <p class="title">Table V</p>
    1064 <div class="caption"><p id="idp547456">XML Namespace Example</p></div>
     1120<p class="title">Table VI</p>
     1121<div class="caption"><p id="idp576096">XML Namespace Example</p></div>
    10651122<table class="table" xml:id="namespace-ex">
    10661123<colgroup span="1">
     
    10961153</table>
    10971154</div>
    1098 <p id="idp556448"> In both Xerces and icXML, every URI has a one-to-one mapping to a URI ID. These
     1155<p id="idp585136"> In both Xerces and icXML, every URI has a one-to-one mapping to a URI ID. These
    10991156            persist for the lifetime of the application through the use of a global URI pool. Xerces
    11001157            maintains a stack of namespace scopes that is pushed (popped) every time a start tag
     
    11041161            those that declare a set of namespaces upfront and never change them, and (2) those that
    11051162            repeatedly modify the namespaces in predictable patterns. </p>
    1106 <p id="idp557584"> For that reason, icXML contains an independent namespace stack and utilizes bit
     1163<p id="idp586272"> For that reason, icXML contains an independent namespace stack and utilizes bit
    11071164            vectors to cheaply perform
    11081165             When a prefix is
     
    11111168            process) to the URI. Each unique namespace binding has a unique namespace id (NSID) and
    11121169            every prefix contains a bit vector marking every NSID that has ever been associated with
    1113               it within the document. For example, in <a class="xref" href="#namespace-ex">Table V</a>, the prefix binding
     1170              it within the document. For example, in <a class="xref" href="#namespace-ex">Table VI</a>, the prefix binding
    11141171            set of <code class="code">p</code> and <code class="code">xmlns</code> would be <code class="code">01</code> and
    11151172            <code class="code">11</code> respectively. To resolve the in-scope namespace binding for each prefix,
     
    11171174            the prefix bit vector with the currently visible namespaces, the in-scope NSID can be
    11181175            found using a bit-scan intrinsic. A namespace binding table, similar to
    1119             <a class="xref" href="#namespace-binding">Table VI</a>, provides the actual URI ID. </p>
     1176            <a class="xref" href="#namespace-binding">Table VII</a>, provides the actual URI ID. </p>
    11201177<div class="table-wrapper" id="namespace-binding">
    1121 <p class="title">Table VI</p>
    1122 <div class="caption"><p id="idp564112">Namespace Binding Table Example</p></div>
     1178<p class="title">Table VII</p>
     1179<div class="caption"><p id="idp592944">Namespace Binding Table Example</p></div>
    11231180<table class="table" xml:id="namespace-binding">
    11241181<colgroup span="1">
     
    11611218</table>
    11621219</div>
    1163 <p id="idp580432">
     1220<p id="idp609264">
    11641221           
    11651222           
     
    11671224           
    11681225         </p>
    1169 <p id="idp582336"> To ensure that scoping rules are adhered to, whenever a start tag is encountered,
     1226<p id="idp611168"> To ensure that scoping rules are adhered to, whenever a start tag is encountered,
    11701227            any modification to the currently visible namespaces is calculated and stored within a
    11711228            stack of bit vectors denoting the locally modified namespace bindings. When an end tag
     
    11781235<div class="section" id="errorhandling">
    11791236<h3 class="title" style="clear: both">Error Handling</h3>
    1180 <p id="idp584768">
     1237<p id="idp613600">
    11811238           
    11821239            Xerces outputs error messages in two ways: through the programmer API and as thrown
     
    11871244            <a class="xref" href="#icxml-arch" title="icXML Architecture">Figure 2</a>, icXML is divided into two sections: the Parabix Subsystem and
    11881245            Markup Processor, each with its own system for detecting and producing error messages. </p>
    1189 <p id="idp587264"> Within the Parabix Subsystem, all computations are performed in parallel, a block at
     1246<p id="idp616160"> Within the Parabix Subsystem, all computations are performed in parallel, a block at
    11901247            a time. Errors are derived as artifacts of bitstream calculations, with a 1-bit marking
    11911248            the byte-position of an error within a block, and the type of error is determined by the
     
    12201277            detected, the sum of those skipped positions is subtracted from the distance to
    12211278            determine the actual column number. </p>
    1222 <p id="idp592784"> The Markup Processor is a state-driven machine. As such, error detection within it
     1279<p id="idp621680"> The Markup Processor is a state-driven machine. As such, error detection within it
    12231280            is very similar to Xerces. However, reporting the correct line/column is a much more
    12241281            difficult problem. The Markup Processor parses the content stream, which is a series of
     
    12361293<div class="section" id="multithread">
    12371294<h2 class="title" style="clear: both">Multithreading with Pipeline Parallelism</h2>
    1238 <p id="idp596320"> As discussed in section <a class="xref" href="#background-xerces" title="Xerces C++ Structure">section “Xerces C++ Structure”</a>, Xerces can be considered a FSM
     1295<p id="idp625216"> As discussed in section <a class="xref" href="#background-xerces" title="Xerces C++ Structure">section “Xerces C++ Structure”</a>, Xerces can be considered a FSM
    12391296         application. These are "embarrassingly
    1240          sequential."<a class="xref" id="idp597472" href="javascript:showcite('cite-Asanovic-EECS-2006-183','idp597472')">Asanovic and others 2006</a> and notoriously difficult to
     1297         sequential."<a class="xref" id="idp626368" href="javascript:showcite('cite-Asanovic-EECS-2006-183','idp626368')">[Asanovic et al. 2006]</a> and notoriously difficult to
    12411298         parallelize. However, icXML is designed to organize processing into logical layers. In
    12421299         particular, layers within the Parabix Subsystem are designed to operate over significant
     
    12441301         well into the general model of pipeline parallelism, in which each thread is in charge of a
    12451302         single module or group of modules. </p>
    1246 <p id="idp598784"> The most straightforward division of work in icXML is to separate the Parabix Subsystem
     1303<p id="idp627680"> The most straightforward division of work in icXML is to separate the Parabix Subsystem
    12471304         and the Markup Processor into distinct logical layers into two separate stages. The
    12481305         resultant application, <span class="ital">icXML-p</span>, is a course-grained
     
    12651322            <code class="code">T<sub>2</sub></code> to finish reading the shared data before it can
    12661323         reuse the memory space. </p>
    1267 <p id="idp609952">
     1324<p id="idp638752">
    12681325        <div class="figure" id="threads_timeline1">
    12691326<p class="title">Figure 4: Thread Balance in Two-Stage Pipelines: Stage 1 Dominant</p>
    1270 <div class="figure-contents"><div class="mediaobject" id="idp611280"><img alt="png image (threads_timeline1.png)" src="threads_timeline1.png" width="500cm"></div></div>
     1327<div class="figure-contents"><div class="mediaobject" id="idp640080"><img alt="png image (threads_timeline1.png)" src="threads_timeline1.png" width="500cm"></div></div>
    12711328</div>
    12721329        <div class="figure" id="threads_timeline2">
    12731330<p class="title">Figure 5: Thread Balance in Two-Stage Pipelines: Stage 2 Dominant</p>
    1274 <div class="figure-contents"><div class="mediaobject" id="idp614336"><img alt="png image (threads_timeline2.png)" src="threads_timeline2.png" width="500cm"></div></div>
     1331<div class="figure-contents"><div class="mediaobject" id="idp643088"><img alt="png image (threads_timeline2.png)" src="threads_timeline2.png" width="500cm"></div></div>
    12751332</div>
    12761333      </p>
    1277 <p id="idp616368"> Overall, our design is intended to benefit a range of applications. Conceptually, we
     1334<p id="idp645120"> Overall, our design is intended to benefit a range of applications. Conceptually, we
    12781335         consider two design points. The first, the parsing performed by the Parabix Subsystem
    12791336         dominates at 67% of the overall cost, with the cost of application processing (including
     
    12811338         scenario, the cost of application processing dominates at 60%, while the cost of XML
    12821339         parsing represents an overhead of 40%. </p>
    1283 <p id="idp617280"> Our design is predicated on a goal of using the Parabix framework to achieve a 50% to
     1340<p id="idp646032"> Our design is predicated on a goal of using the Parabix framework to achieve a 50% to
    12841341         100% improvement in the parsing engine itself. In a best case scenario, a 100% improvement
    12851342         of the Parabix Subsystem for the design point in which XML parsing dominates at 67% of the
     
    12891346         about 33% of the original work. In this case, Amdahl's law predicts that we could expect up
    12901347         to a 3x speedup at best. </p>
    1291 <p id="idp618400"> At the other extreme of our design range, we consider an application in which core
     1348<p id="idp647152"> At the other extreme of our design range, we consider an application in which core
    12921349         parsing cost is 40%. Assuming the 2x speedup of the Parabix Subsystem over the
    12931350         corresponding Xerces core, single-threaded icXML delivers a 25% speedup. However, the most
     
    12951352         the entire latency of parsing within the serial time required by the application. In this
    12961353         case, we achieve an overall speedup in processing time by 1.67x. </p>
    1297 <p id="idp619344"> Although the structure of the Parabix Subsystem allows division of the work into
     1354<p id="idp648096"> Although the structure of the Parabix Subsystem allows division of the work into
    12981355         several pipeline stages and has been demonstrated to be effective for four pipeline stages
    1299          in a research prototype <a class="xref" id="idp619824" href="javascript:showcite('cite-HPCA2012','idp619824')">Lin and Medforth 2012</a>, our analysis here suggests that the further
     1356         in a research prototype <a class="xref" id="idp648576" href="javascript:showcite('cite-HPCA2012','idp648576')">[Lin and Medforth 2012]</a>, our analysis here suggests that the further
    13001357         pipelining of work within the Parabix Subsystem is not worthwhile if the cost of
    13011358         application logic is little as 33% of the end-to-end cost using Xerces. To achieve benefits
     
    13051362<div class="section" id="performance">
    13061363<h2 class="title" style="clear: both">Performance</h2>
    1307 <p id="idp622208"> We evaluate Xerces-C++ 3.1.1, icXML, icXML-p against two benchmarking applications: the
     1364<p id="idp650960"> We evaluate Xerces-C++ 3.1.1, icXML, icXML-p against two benchmarking applications: the
    13081365         Xerces C++ SAXCount sample application, and a real world GML to SVG transformation
    13091366         application. We investigated XML parser performance using an Intel Core i7 quad-core (Sandy
     
    13111368         L1 cache, 256 kB (per core) L2 cache, 8 MB L3 cache) running the 64-bit version of Ubuntu
    13121369         12.04 (Linux). </p>
    1313 <p id="idp623120"> We analyzed the execution profiles of each XML parser using the performance counters
     1370<p id="idp651872"> We analyzed the execution profiles of each XML parser using the performance counters
    13141371         found in the processor. We chose several key hardware events that provide insight into the
    13151372         profile of each application and indicate if the processor is doing useful work. The set of
    13161373         events included in our study are: processor cycles, branch instructions, branch
    13171374         mispredictions, and cache misses. The Performance Application Programming Interface (PAPI)
    1318          Version 5.5.0 <a class="xref" id="idp623888" href="javascript:showcite('cite-papi','idp623888')">Innovative Computing Laboratory</a> toolkit was installed on the test system to facilitate the
     1375         Version 5.5.0 <a class="xref" id="idp652640" href="javascript:showcite('cite-papi','idp652640')">[PAPI]</a> toolkit was installed on the test system to facilitate the
    13191376         collection of hardware performance monitoring statistics. In addition, we used the Linux
    1320          perf <a class="xref" id="idp624816" href="javascript:showcite('cite-perf','idp624816')">Eranian and Gouriou</a> utility to collect per core hardware events. </p>
    1321 <div class="section" id="idp625760">
     1377         perf <a class="xref" id="idp653568" href="javascript:showcite('cite-perf','idp653568')">[perf]</a> utility to collect per core hardware events. </p>
     1378<div class="section" id="idp654464">
    13221379<h3 class="title" style="clear: both">Xerces C++ SAXCount</h3>
    1323 <p id="idp626400"> Xerces comes with sample applications that demonstrate salient features of the
     1380<p id="idp655104"> Xerces comes with sample applications that demonstrate salient features of the
    13241381            parser. SAXCount is the simplest such application: it counts the elements, attributes
    13251382            and characters of a given XML file using the (event based) SAX API and prints out the
    13261383            totals. </p>
    1327 <p id="idp627104"> <a class="xref" href="#XMLdocs">Table VII</a> shows the document characteristics of the XML input files
     1384<p id="idp655808"> <a class="xref" href="#XMLdocs">Table VIII</a> shows the document characteristics of the XML input files
    13281385            selected for the Xerces C++ SAXCount benchmark. The jaw.xml represents document-oriented
    13291386            XML inputs and contains the three-byte and four-byte UTF-8 sequence required for the
     
    13311388            documents and consist entirely of single byte encoded ASCII characters.
    13321389  <div class="table-wrapper" id="XMLdocs">
    1333 <p class="title">Table VII</p>
    1334 <div class="caption"><p id="idp629440">XML Document Characteristics</p></div>
     1390<p class="title">Table VIII</p>
     1391<div class="caption"><p id="idp658256">XML Document Characteristics</p></div>
    13351392<table class="table" xml:id="XMLdocs">
    13361393<colgroup span="1">
     
    13811438</div>           
    13821439</p>
    1383 <p id="idp644944"> A key predictor of the overall parsing performance of an XML file is markup
    1384            density<sup class="fn-label"><a href="#idp645312" class="footnoteref" id="idp645312-ref">[2]</a></sup>. This metric has substantial influence on the
     1440<p id="idp673856"> A key predictor of the overall parsing performance of an XML file is markup
     1441           density<sup class="fn-label"><a href="#idp674224" class="footnoteref" id="idp674224-ref">[2]</a></sup>. This metric has substantial influence on the
    13851442            performance of traditional recursive descent XML parsers because it directly corresponds
    13861443            to the number of state transitions that occur when parsing a document. We use a mixture
    13871444            of document-oriented and data-oriented XML files to analyze performance over a spectrum
    13881445            of markup densities. </p>
    1389 <p id="idp646480"> <a class="xref" href="#perf_SAX" title="SAXCount Performance Comparison">Figure 6</a> compares the performance of Xerces, icXML and pipelined icXML
     1446<p id="idp675392"> <a class="xref" href="#perf_SAX" title="SAXCount Performance Comparison">Figure 6</a> compares the performance of Xerces, icXML and pipelined icXML
    13901447            in terms of CPU cycles per byte for the SAXCount application. The speedup for icXML over
    13911448            Xerces is 1.3x to 1.8x. With two threads on the multicore machine, icXML-p can achieve
     
    13941451            icXML-p performs better as markup-density increases because the work performed by each
    13951452            stage is well balanced in this application. </p>
    1396 <p id="idp648224">
     1453<p id="idp677136">
    13971454        <div class="figure" id="perf_SAX">
    13981455<p class="title">Figure 6: SAXCount Performance Comparison</p>
    13991456<div class="figure-contents">
    1400 <div class="mediaobject" id="idp649536"><img alt="png image (perf_SAX.png)" src="perf_SAX.png" width="500cm"></div>
     1457<div class="mediaobject" id="idp678448"><img alt="png image (perf_SAX.png)" src="perf_SAX.png" width="500cm"></div>
    14011458<div class="caption"></div>
    14021459</div>
     
    14041461         </p>
    14051462</div>
    1406 <div class="section" id="idp652080">
     1463<div class="section" id="idp680992">
    14071464<h3 class="title" style="clear: both">GML2SVG</h3>
    1408 <p id="idp652752">       As a more substantial application of XML processing, the GML-to-SVG (GML2SVG) application
     1465<p id="idp681664">       As a more substantial application of XML processing, the GML-to-SVG (GML2SVG) application
    14091466was chosen.   This application transforms geospatially encoded data represented using
    1410 an XML representation in the form of Geography Markup Language (GML) <a class="xref" id="idp653280" href="javascript:showcite('cite-lake2004geography','idp653280')">Lake and Burggraf 2004</a>
     1467an XML representation in the form of Geography Markup Language (GML) <a class="xref" id="idp682192" href="javascript:showcite('cite-lake2004geography','idp682192')">[Lake and Burggraf 2004]</a>
    14111468into a different XML format  suitable for displayable maps:
    1412 Scalable Vector Graphics (SVG) format<a class="xref" id="idp654128" href="javascript:showcite('cite-lu2007advances','idp654128')">Lu and Dos Santos 2007</a>. In the GML2SVG benchmark, GML feature elements
     1469Scalable Vector Graphics (SVG) format <a class="xref" id="idp683088" href="javascript:showcite('cite-lu2007advances','idp683088')">[Lu and Dos Santos 2007]</a>. In the GML2SVG benchmark, GML feature elements
    14131470and GML geometry elements tags are matched. GML coordinate data are then extracted
    14141471and transformed to the corresponding SVG path data encodings.
     
    14181475a known XML format for the purpose of analysis and restructuring to meet
    14191476the requirements of an alternative format.</p>
    1420 <p id="idp655552">Our GML to SVG data translations are executed on GML source data
     1477<p id="idp684464">Our GML to SVG data translations are executed on GML source data
    14211478modelling the city of Vancouver, British Columbia, Canada.
    14221479The GML source document set
     
    14281485<p class="title">Figure 7: Performance Comparison for GML2SVG</p>
    14291486<div class="figure-contents">
    1430 <div class="mediaobject" id="idp657552"><img alt="png image (Throughput.png)" src="Throughput.png" width="500cm"></div>
     1487<div class="mediaobject" id="idp686464"><img alt="png image (Throughput.png)" src="Throughput.png" width="500cm"></div>
    14311488<div class="caption"></div>
    14321489</div>
    14331490</div>
    1434 <p id="idp659888"><a class="xref" href="#perf_GML2SVG" title="Performance Comparison for GML2SVG">Figure 7</a> compares the performance of the GML2SVG application linked against
     1491<p id="idp688752"><a class="xref" href="#perf_GML2SVG" title="Performance Comparison for GML2SVG">Figure 7</a> compares the performance of the GML2SVG application linked against
    14351492the Xerces, icXML and icXML-p.   
    14361493On the GML workload with this application, single-thread icXML
     
    14391496Using icXML-p, a further throughput increase to 111 MB/sec was recorded,
    14401497approximately a 2X speedup.</p>
    1441 <p id="idp661296">An important aspect of icXML is the replacement of much branch-laden
     1498<p id="idp690160">An important aspect of icXML is the replacement of much branch-laden
    14421499sequential code inside Xerces with straight-line SIMD code using far
    14431500fewer branches.  <a class="xref" href="#branchmiss_GML2SVG" title="Comparative Branch Misprediction Rate">Figure 8</a> shows the corresponding
     
    14501507<p class="title">Figure 8: Comparative Branch Misprediction Rate</p>
    14511508<div class="figure-contents">
    1452 <div class="mediaobject" id="idp664080"><img alt="png image (BM.png)" src="BM.png" width="500cm"></div>
     1509<div class="mediaobject" id="idp692896"><img alt="png image (BM.png)" src="BM.png" width="500cm"></div>
    14531510<div class="caption"></div>
    14541511</div>
    14551512</div>
    1456 <p id="idp666368">The behaviour of the three versions with respect to L1 cache misses per kB is shown
     1513<p id="idp695184">The behaviour of the three versions with respect to L1 cache misses per kB is shown
    14571514in <a class="xref" href="#cachemiss_GML2SVG" title="Comparative Cache Miss Rate">Figure 9</a>.   Improvements are shown in both instruction-
    14581515and data-cache performance with the improvements in instruction-cache
     
    14661523<p class="title">Figure 9: Comparative Cache Miss Rate</p>
    14671524<div class="figure-contents">
    1468 <div class="mediaobject" id="idp669168"><img alt="png image (CM.png)" src="CM.png" width="500cm"></div>
     1525<div class="mediaobject" id="idp697984"><img alt="png image (CM.png)" src="CM.png" width="500cm"></div>
    14691526<div class="caption"></div>
    14701527</div>
    14711528</div>
    1472 <p id="idp671456">One caveat with this study is that the GML2SVG application did not exhibit
     1529<p id="idp700272">One caveat with this study is that the GML2SVG application did not exhibit
    14731530a relative balance of processing between application code and Xerces library
    14741531code reaching the 33% figure.  This suggests that for this application and
     
    14801537<div class="section" id="conclusion">
    14811538<h2 class="title" style="clear: both">Conclusion and Future Work</h2>
    1482 <p id="idp673616"> This paper is the first case study documenting the significant performance benefits
     1539<p id="idp702432"> This paper is the first case study documenting the significant performance benefits
    14831540         that may be realized through the integration of parallel bitstream technology into existing
    14841541         widely-used software libraries. In the case of the Xerces-C++ XML parser, the combined
     
    14901547         technologies, this is an important case study demonstrating the general feasibility of
    14911548         these techniques. </p>
    1492 <p id="idp674896"> The further development of icXML to move beyond 2-stage pipeline parallelism is
     1549<p id="idp703712"> The further development of icXML to move beyond 2-stage pipeline parallelism is
    14931550         ongoing, with realistic prospects for four reasonably balanced stages within the library.
    14941551         For applications such as GML2SVG which are dominated by time spent on XML parsing, such a
    14951552         multistage pipelined parsing library should offer substantial benefits. </p>
    1496 <p id="idp675664"> The example of XML parsing may be considered prototypical of finite-state machines
     1553<p id="idp704480"> The example of XML parsing may be considered prototypical of finite-state machines
    14971554         applications which have sometimes been considered "embarassingly
    14981555         sequential" and so difficult to parallelize that "nothing
     
    15001557         point in making the case that parallelization can indeed be helpful across a broad array of
    15011558         application types. </p>
    1502 <p id="idp677040"> To overcome the software engineering challenges in applying parallel bitstream
     1559<p id="idp705856"> To overcome the software engineering challenges in applying parallel bitstream
    15031560         technology to existing software systems, it is clear that better library and tool support
    15041561         is needed. The techniques used in the implementation of icXML and documented in this paper
     
    15071564      </p>
    15081565</div>
    1509 <div class="bibliography" id="idp678048">
     1566<div class="bibliography" id="idp706864">
    15101567<h2 class="title" style="clear:both">Bibliography</h2>
    1511 <p class="bibliomixed" id="CameronHerdyLin2008"><a href="#idp451088">[Cameron and Herdy 2008] </a>Cameron, Robert D., Herdy, Kenneth S. and Lin, Dan. High performance XML parsing using parallel bit stream technology. CASCON'08: Proc. 2008 conference of the center for advanced studies on collaborative research. 2008 New York, NY, USA</p>
    1512 <p class="bibliomixed" id="papi"><a href="#idp623888">[Innovative Computing Laboratory] </a>Innovative Computing Laboratory, University of Texas. Performance Application Programming Interface.<a href="http://icl.cs.utk.edu/papi/" class="link" target="_new">http://icl.cs.utk.edu/papi/</a></p>
    1513 <p class="bibliomixed" id="perf"><a href="#idp624816">[Eranian and Gouriou] </a>Eranian, Stephane, Gouriou, Eric, Moseley, Tipp and Bruijn, Willem de. Linux kernel profiling with perf.<a href="https://perf.wiki.kernel.org/index.php/Tutorial" class="link" target="_new">https://perf.wiki.kernel.org/index.php/Tutorial</a></p>
    1514 <p class="bibliomixed" id="Cameron2008"><a href="#idp450080">[Cameron 2008] </a>Cameron, Robert D.. A case study in SIMD text processing with parallel bit streams: UTF-8 to UTF-16 transcoding. Proc. 13th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming. 2008 New York, NY, USA</p>
    1515 <p class="bibliomixed" id="ParaDOM2009"><a href="#idp302032">[Shah and Rao 2009] </a>Shah, Bhavik, Rao, Praveen, Moon, Bongki and Rajagopalan, Mohan. A Data Parallel Algorithm for XML DOM Parsing. Database and XML Technologies. 2009</p>
    1516 <p class="bibliomixed" id="XMLSSE42"><a href="#idp305184">[Lei 2008] </a>Lei, Zhai. XML Parsing Accelerator with Intel Streaming SIMD Extensions 4 (Intel SSE4). 2008<a href="Intel%20Software%20Network" class="link" target="_new">Intel Software Network</a></p>
    1517 <p class="bibliomixed" id="Cameron2009"><a href="#idp306096">[Cameron and Herdy 2009] </a>Cameron, Rob, Herdy, Ken and Amiri, Ehsan Amiri. Parallel Bit Stream Technology as a Foundation for XML Parsing Performance. Int'l Symposium on Processing XML Efficiently: Overcoming Limits on Space, Time, or Bandwidth. 2009</p>
    1518 <p class="bibliomixed" id="HilewitzLee2006"><a href="#idp523840">[Hilewitz and Lee 2006] </a>Hilewitz, Yedidya and Lee, Ruby B.. Fast Bit Compression and Expansion with Parallel Extract and Parallel Deposit Instructions. ASAP '06: Proc. IEEE 17th Int'l Conference on Application-specific Systems, Architectures and Processors. 2006 Washington, DC, USA</p>
    1519 <p class="bibliomixed" id="Asanovic-EECS-2006-183"><a href="#idp597472">[Asanovic and others 2006] </a>Asanovic, Krste and others. The Landscape of Parallel Computing Research: A View from Berkeley. 2006</p>
    1520 <p class="bibliomixed" id="GRID2006"><a href="#idp286848">[Lu and Chiu 2006] </a>Lu, Wei, Chiu, Kenneth and Pan, Yinfei. A Parallel Approach to XML Parsing. Proceedings of the 7th IEEE/ACM International Conference on Grid Computing. 2006 Washington, DC, USA</p>
    1521 <p class="bibliomixed" id="cameron-EuroPar2011"><a href="#idp306848">[Cameron and Amiri 2011] </a>Cameron, Robert D., Amiri, Ehsan, Herdy, Kenneth S., Lin, Dan, Shermer, Thomas C. and Popowich, Fred P.. Parallel Scanning with Bitstream Addition: An XML Case Study. Euro-Par 2011, LNCS 6853, Part II. 2011 Berlin, Heidelberg</p>
    1522 <p class="bibliomixed" id="HPCA2012"><a href="#idp304240">[Lin and Medforth 2012] </a>Lin, Dan, Medforth, Nigel, Herdy, Kenneth S., Shriraman, Arrvindh and Cameron, Rob. Parabix: Boosting the efficiency of text processing on commodity processors. International Symposium on High-Performance Computer Architecture. 2012 Los Alamitos, CA, USA</p>
    1523 <p class="bibliomixed" id="HPCC2011"><a href="#idp301200">[You and Wang 2011] </a>You, Cheng-Han and Wang, Sheng-De. A Data Parallel Approach to XML Parsing and Query. 10th IEEE International Conference on High Performance Computing and Communications. 2011 Los Alamitos, CA, USA</p>
    1524 <p class="bibliomixed" id="E-SCIENCE2007"><a href="#idp299760">[Pan and Zhang 2007] </a>Pan, Yinfei, Zhang, Ying, Chiu, Kenneth and Lu, Wei. Parallel XML Parsing Using Meta-DFAs. International Conference on e-Science and Grid Computing. 2007 Los Alamitos, CA, USA</p>
    1525 <p class="bibliomixed" id="ICWS2008"><a href="#idp302896">[Pan and Zhang 2008] </a>Pan, Yinfei, Zhang, Ying and Chiu, Kenneth. Hybrid Parallelism for XML SAX Parsing. IEEE International Conference on Web Services. 2008 Los Alamitos, CA, USA</p>
    1526 <p class="bibliomixed" id="IPDPS2008"><a href="#idp300352">[Pan and Zhang 2008] </a>Pan, Yinfei, Zhang, Ying and Chiu, Kenneth. Simultaneous transducers for data-parallel XML parsing. International Parallel and Distributed Processing Symposium. 2008 Los Alamitos, CA, USA</p>
    1527 <p class="bibliomixed" id="HackersDelight"><a href="#idp522832">[Warren 2002] </a>Warren, Henry S.. Hacker's Delight. 2002</p>
    1528 <p class="bibliomixed" id="lu2007advances"><a href="#idp654128">[Lu and Dos Santos 2007] </a>Lu, C.T., Dos Santos, R.F., Sripada, L.N. and Kou, Y.. Advances in GML for geospatial applications. 2007</p>
    1529 <p class="bibliomixed" id="lake2004geography"><a href="#idp653280">[Lake and Burggraf 2004] </a>Lake, R., Burggraf, D.S., Trninic, M. and Rae, L.. Geography mark-up language (GML) [foundation for the geo-web]. 2004</p>
     1568<p class="bibliomixed" id="CameronHerdyLin2008"><a href="#idp453984">[[Parabix1 2008]] </a>Cameron, Robert D., Herdy, Kenneth S. and Lin, Dan. High performance XML parsing using parallel bit stream technology. CASCON'08: Proc. 2008 conference of the center for advanced studies on collaborative research. 2008 New York, NY, USA</p>
     1569<p class="bibliomixed" id="papi"><a href="#idp652640">[[PAPI]] </a>Innovative Computing Laboratory, University of Texas. Performance Application Programming Interface.<a href="http://icl.cs.utk.edu/papi/" class="link" target="_new">http://icl.cs.utk.edu/papi/</a></p>
     1570<p class="bibliomixed" id="perf"><a href="#idp653568">[[perf]] </a>Eranian, Stephane, Gouriou, Eric, Moseley, Tipp and Bruijn, Willem de. Linux kernel profiling with perf.<a href="https://perf.wiki.kernel.org/index.php/Tutorial" class="link" target="_new">https://perf.wiki.kernel.org/index.php/Tutorial</a></p>
     1571<p class="bibliomixed" id="Cameron2008"><a href="#idp453088">[[u8u16 2008]] </a>Cameron, Robert D.. A case study in SIMD text processing with parallel bit streams: UTF-8 to UTF-16 transcoding. Proc. 13th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming. 2008 New York, NY, USA</p>
     1572<p class="bibliomixed" id="ParaDOM2009"><a href="#idp303712">[[Shah and Rao 2009]] </a>Shah, Bhavik, Rao, Praveen, Moon, Bongki and Rajagopalan, Mohan. A Data Parallel Algorithm for XML DOM Parsing. Database and XML Technologies. 2009</p>
     1573<p class="bibliomixed" id="XMLSSE42"><a href="#idp306864">[[Lei 2008]] </a>Lei, Zhai. XML Parsing Accelerator with Intel Streaming SIMD Extensions 4 (Intel SSE4). 2008<a href="Intel%20Software%20Network" class="link" target="_new">Intel Software Network</a></p>
     1574<p class="bibliomixed" id="Cameron2009"><a href="#idp307776">[[Balisage 2009]] </a>Cameron, Rob, Herdy, Ken and Amiri, Ehsan Amiri. Parallel Bit Stream Technology as a Foundation for XML Parsing Performance. Int'l Symposium on Processing XML Efficiently: Overcoming Limits on Space, Time, or Bandwidth. 2009</p>
     1575<p class="bibliomixed" id="HilewitzLee2006"><a href="#idp534304">[[Hilewitz and Lee 2006]] </a>Hilewitz, Yedidya and Lee, Ruby B.. Fast Bit Compression and Expansion with Parallel Extract and Parallel Deposit Instructions. ASAP '06: Proc. IEEE 17th Int'l Conference on Application-specific Systems, Architectures and Processors. 2006 Washington, DC, USA</p>
     1576<p class="bibliomixed" id="Asanovic-EECS-2006-183"><a href="#idp626368">[[Asanovic et al. 2006]] </a>Asanovic, Krste and others. The Landscape of Parallel Computing Research: A View from Berkeley. 2006</p>
     1577<p class="bibliomixed" id="GRID2006"><a href="#idp288288">[[Lu and Chiu 2006]] </a>Lu, Wei, Chiu, Kenneth and Pan, Yinfei. A Parallel Approach to XML Parsing. Proceedings of the 7th IEEE/ACM International Conference on Grid Computing. 2006 Washington, DC, USA</p>
     1578<p class="bibliomixed" id="cameron-EuroPar2011"><a href="#idp308528">[[Parabix2 2011]] </a>Cameron, Robert D., Amiri, Ehsan, Herdy, Kenneth S., Lin, Dan, Shermer, Thomas C. and Popowich, Fred P.. Parallel Scanning with Bitstream Addition: An XML Case Study. Euro-Par 2011, LNCS 6853, Part II. 2011 Berlin, Heidelberg</p>
     1579<p class="bibliomixed" id="HPCA2012"><a href="#idp305920">[[Lin and Medforth 2012]] </a>Lin, Dan, Medforth, Nigel, Herdy, Kenneth S., Shriraman, Arrvindh and Cameron, Rob. Parabix: Boosting the efficiency of text processing on commodity processors. International Symposium on High-Performance Computer Architecture. 2012 Los Alamitos, CA, USA</p>
     1580<p class="bibliomixed" id="HPCC2011"><a href="#idp302880">[[You and Wang 2011]] </a>You, Cheng-Han and Wang, Sheng-De. A Data Parallel Approach to XML Parsing and Query. 10th IEEE International Conference on High Performance Computing and Communications. 2011 Los Alamitos, CA, USA</p>
     1581<p class="bibliomixed" id="E-SCIENCE2007"><a href="#idp301312">[[Pan and Zhang 2007]] </a>Pan, Yinfei, Zhang, Ying, Chiu, Kenneth and Lu, Wei. Parallel XML Parsing Using Meta-DFAs. International Conference on e-Science and Grid Computing. 2007 Los Alamitos, CA, USA</p>
     1582<p class="bibliomixed" id="ICWS2008"><a href="#idp304576">[[Pan and Zhang 2008a]] </a>Pan, Yinfei, Zhang, Ying and Chiu, Kenneth. Hybrid Parallelism for XML SAX Parsing. IEEE International Conference on Web Services. 2008 Los Alamitos, CA, USA</p>
     1583<p class="bibliomixed" id="IPDPS2008"><a href="#idp302064">[[Pan and Zhang 2008b]] </a>Pan, Yinfei, Zhang, Ying and Chiu, Kenneth. Simultaneous transducers for data-parallel XML parsing. International Parallel and Distributed Processing Symposium. 2008 Los Alamitos, CA, USA</p>
     1584<p class="bibliomixed" id="HackersDelight"><a href="#idp533408">[[Warren 2002]] </a>Warren, Henry S.. Hacker's Delight. 2002</p>
     1585<p class="bibliomixed" id="lu2007advances"><a href="#idp683088">[[Lu and Dos Santos 2007]] </a>Lu, C.T., Dos Santos, R.F., Sripada, L.N. and Kou, Y.. Advances in GML for geospatial applications. 2007</p>
     1586<p class="bibliomixed" id="lake2004geography"><a href="#idp682192">[[Lake and Burggraf 2004]] </a>Lake, R., Burggraf, D.S., Trninic, M. and Rae, L.. Geography mark-up language (GML) [foundation for the geo-web]. 2004</p>
    15301587</div>
    15311588<div class="footnotes">
     
    15341591            behaviour is defined by the inputs, current state and the events associated with
    15351592              transitions of states.</p></div>
    1536 <div id="idp645312" class="footnote"><p><sup class="fn-label"><a href="#idp645312-ref" class="footnoteref">[2]</a></sup> Markup Density: the ratio of markup bytes used to define the structure
     1593<div id="idp674224" class="footnote"><p><sup class="fn-label"><a href="#idp674224-ref" class="footnoteref">[2]</a></sup> Markup Density: the ratio of markup bytes used to define the structure
    15371594             of the document vs. its file size.</p></div>
    15381595</div>
Note: See TracChangeset for help on using the changeset viewer.