Workflow
•
Random DAGs
The random DAGs have been generated with
GGen and could be download through this link: Random DAGs.
Description
Each DAG is described in the dot format. The i th DAG
structure is provided in the file 'dag_$i.dot'.
In these files, there is only the DAG structure.
Every required information about the application
are provided in the file 'appli_$i.dot'.
On each node the size parameter provides the amount of work
(in flops)
required to complete the task.
On each edge the size parameter provides the amount of data
(in bits)
exchanged between two tasks.
The file 'appli_$i.dot.wscom' provides
the DAG with fork task used by WSCOM.
The file with the prefix 'analyse' details some
characteristics about the application.
The way used to generate the files, could be found
in files with the prefix 'stderr'.
The set of files is split into four folders. Each folder represents one
parameter set for the DAG structure generation. In these
folders there are also four folders. The folder name defines
the maximal amount of data which is exchanged between two
tasks.
• MacTrA (Macport TRace Archive) This Workflow has been generated from some package include in MacPort. MacTrA is the DAG application archive used to compare WSCOM and the classical work-stealing by varying the platform bandwidth. The representation is the same as the one of random DAGs, however there is only one folder. The DAG generation has been done by adding some output inside the software make. This output has been script to generate them. This part is not provided yet. It requires some engineering stuffs to become more friendly user. The execution of each package has been done on the same computer sequentially to avoid any perturbation. This computer was a Mac Pro of 2007 with two 3 GHz Quad Core Xeon X5365 and 6 Go of RAM. The whole computation took around one week.