Playground documentation

diff --git a/doc/resource.rst b/doc/resource.rst
index 8a40e38cc1e4c1683426b8719ff896b160c02dc4..8fa9c9182847641798f45327aa29f7e5ba4b5212 100644 (file)
--- a/doc/resource.rst
+++ b/doc/resource.rst
@@ -1,13 +1,13 @@
 .. _resource:
 
-*******************************************************
+**************************************************
 Concurrent branches and multiple machine execution
-*******************************************************
+**************************************************
 
 .. _concurrent:
 
 Execution of concurrent branches
-===================================
+================================
 YACS can execute calculation nodes of a scheme simultaneously.  
 However, simultaneous execution of a large number of nodes can saturate the system.  
 The maximum number of simultaneous executions can be controlled by fixing the maximum number of threads used with the 
@@ -23,7 +23,7 @@ default value is 1048576 (1MB).
 .. _multi:
 
 Execution on multiple machines
-===================================
+==============================
 YACS can execute the nodes of a scheme on several machines where SALOME is
 already installed.
 Each machine is a resource which has to be declared in the resources catalog
@@ -33,3 +33,24 @@ Every node is executed by a container.
 Containers use a set of constraints and rules for choosing the resource where
 the node will be executed (see :ref:`containers`).
 
+
+Execution on multiple machines with mutliples HP container types
+================================================================
+
+For a complex graph using multiples HP containers it can be really difficult to determine the best values 
+of pool sizes for each HP container type and the numbers of branchs of the ForEach loops. In order to let Yacs resolves 
+this problem by itself, one can use the Playground feature: ::
+
+   pilot.PlayGround()
+   machineCatalog=[('is2541',8),] # [('machine name', 'nb cores on machine'), ...]
+   pg.setData(machineCatalog)
+   p.fitToPlayGround(pg)
+
+Principle:
+The graph is analyzed and recursively, for all fork, the current domain is equitably divided into sub-domains 
+(one for each sub-branch). A domain of a branch corresponds to the cores from the machine catalog which will be used 
+for the execution of the nodes of this branch. This division in sub-domains assures that one core will be used by at 
+most one container at once, even if it can be used by different HP containers type during the whole execution of a graph.
+
+Once all the domains have been set, the numbers of branchs of the ForEach loops are calculated in order to use the whole domain associated in runtime.
+Note: If a branch does not contain any ForEach loop, the size of its domain is reduced to the minimal size.