📄 http:^^www.cs.utexas.edu^users^oops^papers.html
字号:
MIME-Version: 1.0
Server: CERN/3.0
Date: Tuesday, 07-Jan-97 15:53:49 GMT
Content-Type: text/html
Content-Length: 14359
Last-Modified: Friday, 28-Jun-96 03:51:40 GMT
<HTML><head><title>OOPS Group Publications</title></head><body><h1>OOPS Group Publications</h1><ol><li> <a name="rttdshrt"> Sheetal V. Kakkad, Mark S. Johnstone, andPaul R. Wilson. <b>Portable Runtime TypeDescription for Conventional Compilers</b>. Submitted toUSENIX '97.<blockquote>Many useful language extensions and support libraries require knowledge ofthe layout of fields within objects at runtime. Examples include orthogonalpersistent object stores, garbage collectors, data structure picklers, parameter marshalling schemes, etc.<p>For clean and efficient implementation as libraries, these systems requireknowledge of the actual layouts of data objects, which is unavailable inmost traditionally-compiled and linked programming languages, such as C, C++,and Ada.<p>We present a facility for runtime type description, orRTTD, which extracts the low-level layout information from debug output ofconventional compilers, and makes it available to user programs. We describethe basic strategies of the system, and present details of our interface forC++. We also sketch some extensions we have implemented, including specialtreatment of C++'s virtual function table pointers.<p>Our implementation is freely available via anonymous ftp.</blockquote><!WA0><!WA0><!WA0><!WA0><!WA0><a href="ftp://ftp.cs.utexas.edu/pub/garbage/rttdshrt.ps"><!WA1><!WA1><!WA1><!WA1><!WA1><img align=bot src="http://www.cs.utexas.edu/icons/doc.xbm" alt=""></a><!WA2><!WA2><!WA2><!WA2><!WA2><a href="ftp://ftp.cs.utexas.edu/pub/garbage/rttdshrt.ps">Postscript <i>(136KB)</i></a><p><li> <a name=neely-thesis> Michael Neely. <b>An Analysis of theEffects of Memory Allocation Policy on Storage Fragmentation</b>.Master's thesis, The University of Texas at Austin, May 1996. </a> <p><blockquote>The study of dynamic memory allocation has been dominated bymeasurement of performance of allocators with random input streams ofrequests. This study introduces a new methodology that separates theissue of policy from implementation and focuses on the effects ofplacement policy on fragmentation. It studies the effects of policydecisions such as object placement, splitting, and coalescing, onstorage fragmentation. A useful and accurate measurement offragmentation is presented that is based on the amount of waste at thepoint of peak memory usage. We have attempted to pick a representativeset of allocators from the space of reasonable combinations of knownpolicies. The allocators are used in memory allocation simulations todetermine their respective fragmentation.<p>Our results show that the best fit (FIFO, LIFO or address-ordered) andaddress-ordered first fit policies yield the lowest fragmentation onaverage (14%), and that the overheads associated with these allocatorsare the largest source of wasted memory. We also explain how best fitcan be implemented efficiently. A representative set of real programallocation traces is used in the simulations, and compared withrandomized traces, to show that the application's patterns ofallocation are an important factor in the allocator's performance andthat studies based on synthetic traces are fundamentally flawed.</blockquote><!WA3><!WA3><!WA3><!WA3><!WA3><a href="ftp://ftp.cs.utexas.edu/pub/garbage/neely-thesis.ps.gz"><!WA4><!WA4><!WA4><!WA4><!WA4><img align=bot src="http://www.cs.utexas.edu/icons/doc.xbm" alt=""></a><!WA5><!WA5><!WA5><!WA5><!WA5><a href="ftp://ftp.cs.utexas.edu/pub/garbage/neely-thesis.ps.gz">Compressed Postscript <i>(184KB)</i></a><p><li> <a name="allocsrv"> Paul R. Wilson, Mark S. Johnstone, MichaelNeely, and David Boles. <b>Dynamic Storage Allocation:A Survey and Critical Review</b>. In <cite>International Workshop on Memory Management</cite>, Kinross, Scotland, UK, September 1995. </a> <p><blockquote>Dynamic memory allocation has been a fundamental part of most computersystems since roughly 1960, and memory allocation is widelyconsidered to be either a solved problem or an insoluble one.In this survey, we describe a variety of memory allocator designsand point out issues relevant to their design and evaluation. Wethen chronologically survey most of the literature on allocators between1961 and 1995. (Scores of papers are discussed, in varying detail,and over 150 references are given.)<p>We argue that allocator designs have been unduly restricted byan emphasis on mechanism, rather than policy, while the latter ismore important; higher-level <i>strategic</i> issues are stillmore important, but have not been given much attention.<p>Most theoretical analyses and empirical allocator evaluations to date have relied on very strong assumptions of randomness and independence, but real program behavior exhibits important regularities that must beexploited if allocators are to perform well in practice.</blockquote><!WA6><!WA6><!WA6><!WA6><!WA6><a href="ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps"><!WA7><!WA7><!WA7><!WA7><!WA7><img align=bot src="http://www.cs.utexas.edu/icons/doc.xbm" alt=""></a><!WA8><!WA8><!WA8><!WA8><!WA8><a href="ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps">Postscript <i>(923KB)</i></a><p><li> <a name="prefetchsim"> Paul R. Wilson, Sheetal Kakkad, andShubhendu S. Mukherjee. <b>Anomalies and Adaptation in the Analysisand Development of Prefetching Policies</b>. <cite> Journal of Systemsand Software</cite>, 27(2):147-153, November 1994. Technical communication.</a> <p><blockquote>In "Analysis and Development of Demand Prepaging Policies,"Horspool and Huberman show that it is possible todesign prefetching memory policies that preserve a "stack"inclusion property, much like LRU, allowing them to simulatethese policies for all sizes of memory in a single pass througha reference trace. We believe that the details of Horspool andHuberman's algorithms introduce unexpected anomalous properties,however. In particular, their policies are not properly<i>timescale relative</i>--events occuring on a timescalethat should only matter tosome sizes of memory adversely affect replacement decisionsfor memories of very different sizes. Slight changes to thealgorithms can restore timescale relativity and make them muchbetter-behaved. In addition, we would like to point out thatHorspool and Huberman's algorithms actually simulate <i>adaptive</i>policies, which may explain some of their unexpectedly positiveresults. This view suggests that properly timescale-relativeinclusion-preserving policies can be used to systematicallyevaluate adaptive memory management schemes.</blockquote><!WA9><!WA9><!WA9><!WA9><!WA9><a href="ftp://ftp.cs.utexas.edu/pub/garbage/prefetchsim.ps"><!WA10><!WA10><!WA10><!WA10><!WA10><img align=bot src="http://www.cs.utexas.edu/icons/doc.xbm" alt=""></a><!WA11><!WA11><!WA11><!WA11><!WA11><a href="ftp://ftp.cs.utexas.edu/pub/garbage/prefetchsim.ps">Postscript <i>(123KB)</i></a><p><li> <a name="real-time-gc"> Paul R. Wilson and Mark S. Johnstone.<b>Real-Time Non-Copying Garbage Collection</b>. Position paper forthe <cite> 1993 ACM OOPSLA Workshop on Memory Management and GarbageCollection</cite>, Washington D.C., September 1993.</a> <p><!WA12><!WA12><!WA12><!WA12><!WA12><a href="ftp://ftp.cs.utexas.edu/pub/garbage/GC93/wilson.ps"><!WA13><!WA13><!WA13><!WA13><!WA13><img align=bot src="http://www.cs.utexas.edu/icons/doc.xbm" alt=""></a><!WA14><!WA14><!WA14><!WA14><!WA14><a href="ftp://ftp.cs.utexas.edu/pub/garbage/GC93/wilson.ps">Postscript <i>(102KB)</i></a><p>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -