[Mpi3-rma] Next RMA Telecon

Torsten Hoefler htor at illinois.edu
Mon Dec 26 23:15:00 CST 2011

On Mon, Dec 26, 2011 at 10:49:55PM -0600, Pavan Balaji wrote:
> On 12/26/2011 10:45 PM, Torsten Hoefler wrote:
>> No, the MPI library does not have to deal with silly use programs. If a
>> user decides to write a non-scalable program, then the MPI library
>> cannot do much.
> My question is about correctness, not performance scalability.  It can  
> be slow, but it has to work.  Also, if the number of processes is small,  
> it's not really a scalability issue either.
I see, I saw only the scalability issue. Yes, it is a correct program. 

> My point is that, the user can still do this, so the MPI implementation  
> has to deal with it.  So not providing Win_lock_all(EXCLUSIVE) is not  
> really reducing the implementation complexity.  The lock_all call does  
> give the opportunity to make the implementation more scalable (and  
> possibly more complex), but that's a quality of implementation issue.
> So, what am I missing here?
Right, Brian had some arguments why/how this would be either a
non-scalable implementation (in memory and time) or a burden on the
implementer. So we didn't want to enforce to have a good implementation
of this (since our only use-case was solved otherwise). 

Thanks & Best,

 bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ -----
Torsten Hoefler         | Performance Modeling and Simulation Lead
Blue Waters Directorate | University of Illinois (UIUC)
1205 W Clark Street     | Urbana, IL, 61801
NCSA Building           | +01 (217) 244-7736

More information about the mpiwg-rma mailing list