Re: JavaMemoryModel: Isolation properties in the JMM

From: Sylvia Else (
Date: Fri Feb 06 2004 - 08:29:14 EST

At 05:00 AM 6/02/2004, Bill Pugh wrote:

>The idea of the isolation property is for threads not in the same
>partition, you only need to examine their behavior in this execution
>enough to determine that they are not in the same partition.
>Without this property, you may have to consider all potential behaviors of
>all threads, including threads that weren't even created in the execution
>you are trying to understand.

Yet as a developer, I tend not to want to understand executions, so much as
predict them. I rather question the utility of the partioning scheme you
described, or its associated isolation property. I note, though, that it
appears that SC- cannot be partitioned this way.

Even your partitioning model has the characteristic that modifications to
the program that on the fact of it shouldn't affect the isolation
nevertheless bring threads previous separately partitioned into one
partition. An example is where both threads are modified to make reads of a
common variable.

Both models could be partitioned statically, and a static partitioning has
more scope for being used predictively.


JavaMemoryModel mailing list -

This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:57 EDT