送交者: BerkeleyWolf 于 2005-3-11, 11:18:53:
回答: 老狼, thx for the effort 由 sssa 于 2005-3-11, 10:51:53:
First, the 1994 paper set the framework of how to measure MPI, the metrics etc. 2000 paper completely follows and 剽窃. In term of the framework, the 1994 paper is more complete.
Second, yes, they measure different things. The second paper measures a IBM SP2 computer of 28 RS/6000 nodes . However, they only measure Sending Latency, and from a fuzzy image, you even can not understand what are these four lines.
Third, 2000's conclusions are vague and without support. (1) conlclusion 1: software overhead is very high 剽窃. (2) conclusion 2: selecting the buffered mode. There is no discussion of buffered mode before. How this conclusion comes from? (3) conclusion 3: replace the standard functions with sendrecv. What is sendrecv? never mentioned before and can not understand what sendrecv is.
In scientific research, conclusions without support suck!!!!!!!!