Re: JavaMemoryModel: Implementing atomic object creation on Alphas may be cheap

From: Bill Pugh (pugh@cs.umd.edu)
Date: Wed Oct 13 1999 - 11:44:02 EDT


>However, one big problem I don't understand. Sanjay's numbers came out to be
>4,329,317,558 total memory barriers for one run of each program (his
>original numbers reflected 3 runs of each program). That is about 21
>times more memory barriers that we get if we simple instrument every
>getfield.

I found out what the problem was. We were using -s1, which specified
the toy data sets, while Sanjay was using the correct value of -s100,
which gives the full data sets.

We are working on redoing the experiments now. It looks like the
ratios may not be as good for the full data sets, so the technique
I've described may not that useful. I'll let you know as soon as we
get the experiments rerun correctly.

        Bill

-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:19 EDT