📄 http:^^www.cs.wisc.edu^~paradyn^overview.html
字号:
Date: Tue, 05 Nov 1996 20:59:06 GMTServer: NCSA/1.5Content-type: text/htmlLast-modified: Mon, 19 Aug 1996 19:36:02 GMTContent-length: 2782<head><title>Paradyn Project Overview</title></head><body><H1><!WA0><!WA0><!WA0><!WA0><!WA0><IMG SRC="http://www.cs.wisc.edu/~paradyn/paradyn.small.gif" ALT="Paradyn Logo" ALIGN=MIDDLE> Paradyn Project Overview Page</H1><hr><H2>Overview</H1>The Paradyn Parallel Program Performance Tools project is exploring newapproaches to building scalable tools for parallel program performance.<P>Traditional performance tools for sequential machines fail to scale tolarge parallel systems. Normal post-mortem-style tools typicallycollect information about all events that occur during execution; thiscan produce trace data at a rate of several Gigabytes per second on amachine with thousands of nodes, a rate which is impossible for aperformance tool to cope with. The standard solution to this problemis to collect only a small subset of available performance data.However, choosing the right set of metrics is difficult or impossibleto know a priori, and places a large decision-making burden on theprogrammer.<P>Paradyn takes a new approach, called <em>dynamic instrumentation</em>, that is basedon dynamically controlling what performance data is to be collected.Dynamic instrumentation allows data collection instructions to be insertedinto an application program during runtime. Paradynis able to look for a variety of performance problems duringa single execution, while not being overwhelmed with all the availabledata at any one time. In addition, dynamic instrumentation that hasnot be inserted incurs no overhead, in contrast to instrumentationthat is staticly-inserted but currently disabled.<P>Paradyn provides <em>decision support</em> for the tool user. Paradynassists the user in deciding which bottlenecks to look for; in fact,Paradyn can search automatically to isolate performance bottlenecks,and then explain them to the user using descriptions orvisualizations.<p>Paradyn incorporates a new approach for describing performanceinformation to users of high-level parallel languages; Paradyn mapsperformance data between multiple layers of abstraction, and user canchoose to look at it in terms of high-level language constructs orlow-level machine structures.<p>Since Fall 1994, we have been collaborating with the<!WA1><!WA1><!WA1><!WA1><!WA1><a href="http://www.cs.wisc.edu/~wwt/">WisconsinWind Tunnel Project</a>to support Paradyn on their Blizzard platform(both CM-5 and Cluster of Workstation versions).<p>A recent <!WA2><!WA2><!WA2><!WA2><!WA2><a href="http://www.cs.wisc.edu/~wwt/wi_week16nov94.html"><i>Wisconsin Week</i> article</a>summarizes our current ARPA funding and has commentary from Profs. Bart Millerand Mark Hill.<p>Paradyn Release 1.1 is available for SunOS, Solaris, AIX, SP2, HP-UX, andPVM.<p><hr>Last modified:Mon Aug 19 14:35:57 CDT 1996by<!WA3><!WA3><!WA3><!WA3><!WA3><a href="http://www.cs.wisc.edu/~bart">bart</a></b></body>⌨️ 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 + -