📄 http:^^www.cs.utexas.edu^users^code^code-hpcwire-article.html
字号:
MIME-Version: 1.0
Server: CERN/3.0
Date: Friday, 17-Jan-97 00:06:49 GMT
Content-Type: text/html
Content-Length: 10714
Last-Modified: Thursday, 17-Oct-96 20:04:57 GMT
<html> <head> <title>CODE: "Visual Parallel Programming May Come of Age with CODE" (HPCwire)</title> </head> <!-- background="../gracs/background.gif" --> <body background="background.gif" bgcolor="#FFFFFF" LINK="#191eee" VLINK="#b90003" ALINK="#a051b7"><!WA0><!WA0><!WA0><!WA0><a href="http://hpcwire.com/hpcwire.html"><!WA1><!WA1><!WA1><!WA1><img border=0 src="http://www.cs.utexas.edu/users/code/hpc_banner.gif"></a><h2>from <!WA2><!WA2><!WA2><!WA2><a href="http://hpcwire.com/hpcwire.html"><font size=+5 face="Arial, Helvetica">HPCwire</font></a>, 8/23/96:</h2><font size=-1><em>(reprinted with permission)</em></font><p><center> <h1><em>Visual Parallel Programming May Come of Age with CODE</em></h1> <h2><font face="Arial, Helvetica">by Alan Beck, editor-in-chief</font></a></h2></center><center><table width=50%><tr><td><font size=+1>AUSTIN, Texas -- Although visual parallel programming environments are notnew, the unusually simple and direct techniques employed by the University ofTexas' <!WA3><!WA3><!WA3><!WA3><a href="http://www.cs.utexas.edu/users/code">CODE</a> (Computationally Oriented Display Environment), an abstractdeclarative graphical environment for parallel programming, promises not onlya different but also a much more effective approach to the entire task.Within the CODE environment, a parallel program is a directed graph, wheredata flows on arcs connecting nodes representing sequential programs.</font><p> The sequential programs may be written in any language, and CODE itself is architecture-independent. Thus, the CODE system can produce parallel programsfor PVM-based and MPI-based networks as well as for the Sequent Symmetry,CRAY and Sun Sparc MP. Currently, the CODE graphical interface itself runson Suns.<p> In order to learn more about the operation of CODE, HPCwire interviewed its"godfather", <!WA4><!WA4><!WA4><!WA4><a href="http://www.cs.utexas.edu/users/browne">James C. Browne</a>, Regents Chair in Computer Sciences andProfessor of Physics and Electrical & Computer Engineering at the Universityof Texas. Following are selected excerpts from that discussion.<p><hr size=3></td></tr></table><table><tr><td width=20%></td><td><p><b><font color="#0000AF">HPCwire: What level of skill must a programmer bring to CODE?</font></b><p><font color="#000000"> BROWNE: "No special programming skills are needed. Traditionally, oneworks at the procedural level, stipulating how a computation is done. WithCODE, you make a transition from how something is done to what you're tryingto do, at a very high level."<p> "I've used CODE in undergraduate parallel programming classes, givenstudents about an hour of instruction, then turned them loose in thegraphical environment, and they work their way through it. It's forgiving. Ofcourse, they already know how to program in traditional languages. Andclearly, they must understand parallelism at the conceptual level."</font><p><b><font color="#0000AF">HPCwire: Is it fair to say that the better the programmer, the better (s)hewill be able to utilize CODE?</font></b><p><font color="#000000"> BROWNE: "Let's take another tack and say the better people understand theirapplications, the better they'll be able to use CODE. When you write aprogram in a language like HPF, you have to do a lot of nonintuitive thingslike partitioning matrices, handling distributions, etc. But with CODE, whatyou must understand is that a parallel computation structure is basically adirected graph with many paths through it. It's a different level ofconceptual understanding."</font><p></td></tr></table><table><tr><td valign=top width=40%><font size=+2 color='#4A7373'><hr>"I've used CODE in undergraduate parallel programming classes, givenstudents about an hour of instruction, then turned them loose in thegraphical environment, and they work their way through it."<hr></font></td><td><b><font color="#0000AF"> HPCwire: Given a certain level of skill, how much more efficiently can aparallel programmer function with CODE?</font></b><p><font color="#000000"> BROWNE: "An order of magnitude more effectively. And we need to do evenbetter than that. What we've done is change the level of abstraction atwhich programs are described. It's like writing a book by simply writing anoutline, and then having a smart word processor consult encyclopedias anddictionaries to fill in the rest."</font></td></tr></table><table><tr><td width=20%></td><td><p><font color="#000000"> "CODE uses an abstract general model of parallel computation, which is ahierarchical dependence graph. Those are a lot of buzzwords, but thegraphical environment speaks for itself. Let's say you have an old FORTRANprogram with a bunch of subroutines for which you want to invoke manyparallel copies. We associate one subroutine with a node on the graph. If youwant to run <em>n</em> copies of it in parallel and connect them up with the rest ofthe program, you simply draw arcs among the routines you want connected. CODEtakes care of all the parallel programming book-keeping -- like where thecopy is, how to get in touch with it for input or output, how to synchronizeit with other routines."</font><p><b><font color="#0000AF">HPCwire: Doesn't this favor a coarse-grained approach?</font></b><p><font color="#000000"> BROWNE: "Definitely. But about three years ago, we asked ourselves howdata parallelism could be most simply represented in the dependence graphmodel. You can construct a graph where each node represents part of acomputation on a matrix, but that's awkward. So we thought: why notintroduce additional annotation on the arcs that says, 'Any data flowingdown this arc will be split into pieces, with each piece sent to a copy ofthe routine at the arc's end.' If you do this, you then have fine-graineddata parallelism embedded in the graphical model. <!WA5><!WA5><!WA5><!WA5><a href="http://www.cs.utexas.edu/users/code/code-publications.html#ipps96">A paper on this approachto integration of data parallelism into the graphical model</a> was presentedat the last International Parallel Processing Symposium."</font><p><b><font color="#0000AF">HPCwire: What are the limitations of this approach?</font></b><p><font color="#000000"> BROWNE: "Lack of familiarity. The need to change the way you think. See,we're messing with people's minds. There are no intrinsic limitations. Thenodes are nothing more or less than FORTRAN or C subroutines. There are noconceptual limitations on back-ends. We compile to shared-memory, PVM orMPI back-ends. Give us that one graph, click on the icon for the back-endyou want, and we compile it with some optimization for each differentenvironment."<p> "The reason this approach to program development isn't particularlywell-accepted is because the scientific and engineering community isaccustomed to doing business in a certain way -- and working at a certainlevel of abstraction. With this method you require people to change the waythey think. And people change the way they think very slowly."</font><p></td></tr></table><table><tr><td valign=top width=40%><font size=+2 color='#4A7373'><hr>"We compile to shared-memory, PVM orMPI back-ends. Give us that one graph, click on the icon for the back-endyou want, and we compile it with some optimization for each differentenvironment."<hr></font></td><td><b><font color="#0000AF">HPCwire: But how about your competitors?</font></b><p><font color="#000000"> BROWNE: "Many people have worked on graphical models of parallelprogramming for a long time. They have some very good systems. Ted Lewis didthe PPSE system at Oregon State, and there was a group that used to workwith <!WA6><!WA6><!WA6><!WA6><a href="http://www.netlib.org/utk/people/JackDongarra/">Jack Dongarra</a> at the University of Tennessee which produced a systemcalled <!WA7><!WA7><!WA7><!WA7><a href="ftp://ftp.netlib.org/hence/index.html">HeNCE</a> similar to ours. And there are several other interestinggraphical programming systems. We have all encountered similar results --that this was really neat technology, and once you get people trained in itthey like it. But it's like any other major paradigm change -- it's hardfor people to make paradigm shifts."</font></td></tr></table><table><tr><td width=20%></td><td><p><font color="#000000"> "A great part of the benefit is not from the graphical interface --although it makes programming easier -- but from the fact that we're using amore general, more abstract model of parallel computation."</font><p><b><font color="#0000AF">HPCwire: Are you still refining the system?</font></b><p><font color="#000000"> BROWNE: "Yes. We're still trying to better integrate data parallelism withthe dependence-graph model. Also, we're developing a debugger. Normally,debugging parallel programs is very difficult, particularly distributed ones-- because you have race conditions, and you can't wrap your hands around theproblems."<p> "There are really no good debuggers for parallel systems, although somestrong academic work has been done, for example, by Bart Miller at theUniversity of Wisconsin. To debug in CODE we play back the execution byanimating the graph. We can do this because we've cleanly separated thenotion of computation from notions of communication and interaction."</font><p><b><font color="#0000AF">HPCwire: Do you feel this type of model will have a fundamental impact onparallel programming?</font></b><p><font color="#000000"> BROWNE: "I think that over the next 10 years, as people learn to change, aseverybody has a graphical workstation in front of them, the whole idea ofraising the level of abstraction at which people program will take effect. Itruly believe there will be a change in the practice of programming. And bygoing to higher levels of abstraction, there will be more change in the next5 or 10 years than there has been in the 40 years since I wrote my firstprogram."</font><p></td></tr></table></center><hr size=3><center>A free alpha release of CODE is now available. For more information see Website <!WA8><!WA8><!WA8><!WA8><a href="http://www.cs.utexas.edu/users/code/">http://www.cs.utexas.edu/users/code/</a>.</center><p><hr>Copyright 1996 HPCwire. Redistribution of this article is forbidden by law without the expressed written consent of the publisher. For a free trial subscription to HPCwire, send e-mail to <!WA9><!WA9><!WA9><!WA9><a href="http://www.cs.utexas.edu/users/code/trial@hpcwire.tgc.com">trial@hpcwire.tgc.com</a>. <hr><center><!WA10><!WA10><!WA10><!WA10><a href="http://www.cs.utexas.edu/users/code"><!WA11><!WA11><!WA11><!WA11><img border=0 src="http://www.cs.utexas.edu/users/code/code-icon-trim.gif"></a><b><FONT SIZE="-1"><address><!WA12><!WA12><!WA12><!WA12><A HREF="mailto:emery@cs.utexas.edu">emery@cs.utexas.edu</A> / <!-- Created: Tue Oct 8 11:11:41 CDT 1996 --><!-- hhmts start -->Last modified: Tue Oct 8 18:25:25 CDT <!-- hhmts end --></address></FONT></b></center> </body></html>⌨️ 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 + -