Re: JavaMemoryModel: cost of initialization-safety on an Alpha SMP

From: Doug Lea (dl@altair.cs.oswego.edu)
Date: Sat Aug 28 1999 - 20:18:25 EDT


> Bill asked me to collect some more information related to the
> cost of initialization safety.

These results are very helpful. But to me, the most helpful
numbers would be those comparing specJVM runs in which
  All String constructors and accessors are synchronized
vs
  All String constructors and accessors use barriers as needed to
  preserve initialization safety
vs
  current version of String

String is far most prevalent case of a class that must (so far as I
can see anyway) be changed pending outcome of initialization safety
rules. If there is no guaranteed initialization safety at all, then it
seems that Strings must use synch. If the guarantee holds only for
finals (as in my
http://www.cs.umd.edu/~pugh/java/memoryModel/archive/0154.html), then
Strings must use finals, which entails barriers on alpha.

I imagine that String represents the majority of all uses of immutable
objects in SpecJVM programs. So, just measuring impact on class
String would give an adequate perspective.

How hard would it be to collect this data?

-- 
Doug Lea, Computer Science Department, SUNY Oswego, Oswego, NY 13126 USA
dl@cs.oswego.edu 315-341-2688 FAX:315-341-5424 http://gee.cs.oswego.edu/  
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:18 EDT