Re: JavaMemoryModel: Agreement to disagree: where things stand

From: victor.luchangco@sun.com
Date: Tue Jul 29 2003 - 09:39:26 EDT


> OK, does anyone have suggestions for alternative names?
>
> We could just call it the Manson/Pugh model, and say that it consists
> of two properties, consistency or causality.
>
> Or we can try to find synonyms for consistency and causality.

If I understand correctly, the purpose of "causality" is to avoid
"self-causing" operations. So why not just call it "non-self-causing"
or something like that? Since it is just part of the definition of a
single memory model, and not even well-defined by itself (you need the
definition of "consistency" to define causality), I don't think it's
too important to come up with a spiffy name for it. If the CnC model
is adopted, it'll just be called the Java memory model, the JMM, or
the Manson/Pugh model anyway.

As for "consistency", that model is essentially the dag consistency
model first proposed by Blumofe et al. (IPPS 96), where the dag has
the actions as nodes and the "happens before" relation as edges. So
why not just call it that (or more precisely, WW-dag consistency)?
Alternatively, if that name is too obscure, it could be called
"happens before" consistency, or more accurately, "allowable reads"
consistency. Those are kludgy names, but again, they won't be used
much outside the model definition, and the notions of "happens before"
and "allowable reads" are already used heavily within the definition.

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



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:47 EDT