[Prev][Next][Index][Thread]

[kathy@TIDE.LCS.MIT.EDU: practice talk]



Date: Thu, 26 May 88 10:39:26 edt
From: kathy@TIDE.LCS.MIT.EDU (Kathy Yelick)
To: spd@xx.LCS.MIT.EDU, tds@xx.LCS.MIT.EDU
Subject: practice talk

I will be giving a practice talk on Friday at 11:15 in room 516.  This
is practice for a 30 minute presentation at the Topaz Users' Group meeting.
Comments on the both the talk and the research will be appreciated.  


	 Problems with Parallel Program Specifications
		    (work in progress)

			    by

			Kathy Yelick

The difficulty of writing and reasoning about parallel programs is one of the
barriers to effectively using multi-processors.  Some specification techniques
use conventional sequential specifications, and require that correct
implementations mimic the sequential behavior in the context of arbitrary
concurrency.  Such methods may require less efficient implementations than
would naturally be chosen by an implementor.  This problem is especially 
important with large-grain parallelism, where parallelism does not exist
at every level of abstraction.  We augment specifications with interference
information, and show that these specifications may allow more efficient 
programs.