Meeting 7 , 8 and 9 [01/06, 13/06 and 20/06]

Meeting 7 [01/06]

During this meeting the project was registered on Sourceforge:

• project page
• And the svn: mpi-visu svn

The meeting was very short and the goal of the next coming week was to develop the user-synchronised communication: literally waiting for the user to click on a continue button to continue the MPI operation.

Meeting 8 [13/06]

During the 8th meeting the result of the synchronised communication with the user was presented. Some improvement were nonetheless needed. Each message between the interface and the profiler are done though formatted string, chosen for the simplicity of adding new elements in it. It also gives the advantage of variable length, as MPI_Probe could be use on the receiver side to determine the size of the message and avoid sending big messages when not required (messages for registering a communicator and giving information about a Ssend are not the same length for instance, as they do not have the same information to transmit).

Meeting 9 [20/06]

During the week the code was cleaned, reorganising the Interface mostly in order to cope with user-waiting communications. Registering a communicator was also completed, so the profiler can cope with adding and removing communicators easily. Nonetheless the attribute to find if the communicator was previously registered wasn't introduced yet.

Some improvements were proposed by David on the way of displaying the communications. Currently a ball is moving from the waiting processor to the waited one (A does a Ssend to B, the ball moves from A to B). But displaying a fix image seams to be a better idea, as a moving ball implies message transit, when there is none.

An error was also found when using the Compute Pi example, as the tag MPI_ANY_SOURCE was used and the Interface couldn't understand it.

The next step is to develop the basis of the array registration. With that done, each main requirement will be completed, forming a backbone to develop further one or the other by taking a specific example. In order to register memory nonetheless some development has to be made on both the profiler and interface, as currently there is no way of correctly finding if a peace of memory was registered and to display it on the interface.