JavaMemoryModel: Relaxing SC to avoid global total orders.

From: Sylvia Else (sylviae@optushome.com.au)
Date: Wed Dec 17 2003 - 18:18:52 EST


Hi,

To quote from Bill and Jeremy's "Proof sketch that Manson/Pugh allows
reordering"...

"We say an execution has sequentially consistent (SC) results if its
results are the same as if
the actions of all the thread were executed in some sequential order, and
the actions of each
individual thread appear in this sequence in program order."

I'd suggest that when a programmer uses SC reasoning, they're actually
thinking in terms of the viewpoint of each thread separately, coupled with
relationships with other threads that are defined by the program's
synchronization actions.

So an alternative model would be:

Each thread of a correctly synchronized program sees all of the actions of
the program occur in some sequential order, and the actions of each
individual thread appear in this sequence in program order. But with the
caveat that the sequential order seen by each thread may differ from that
seen by other threads.

This would eliminate the need for a global total order over synchronization
actions.

Sylvia.

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



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:55 EDT