add random assignment two step rebalancing

tests/test_dataDistribution.py

@@ -259,7 +259,7 @@ class TestDataDistribution(unittest.TestCase):
         distribution.add_osd_list(create_test_osd_list(num_osds, osd_bandwidths))
         distribution.add_folders(create_test_folder_list(num_folders, folder_sizes), random_osd_assignment=True)
 
-        distribution.rebalance_two_steps()
+        distribution.rebalance_two_steps_optimal_matching()
 
         osds = distribution.get_osd_list()
         total_folder_sizes = list(map(lambda x: distribution.OSDs[x].total_folder_size, osds))
@@ -276,7 +276,7 @@ class TestDataDistribution(unittest.TestCase):
         distribution.add_osd_list(create_test_osd_list(num_osds, osd_bandwidths))
         distribution.add_folders(create_test_folder_list(num_folders, folder_sizes), random_osd_assignment=True)
 
-        distribution.rebalance_two_steps()
+        distribution.rebalance_two_steps_optimal_matching()
 
         osds = distribution.get_osd_list()
         total_folder_sizes = list(map(lambda x: distribution.OSDs[x].total_folder_size, osds))

xtreemfs_client/dataDistribution.py

@@ -340,7 +340,7 @@ class DataDistribution(object):
 
         return movements
 
-    def rebalance_two_steps(self):
+    def rebalance_two_steps_optimal_matching(self):
         """
         rebalance the distribution in two steps:
             1. calculate new distribution, independently of the current one
@@ -440,6 +440,31 @@ class DataDistribution(object):
 
         return movements
 
+    def rebalance_two_steps_random_matching(self):
+        """
+        rebalance the distribution in two steps:
+            1. calculate new distribution, independently of the current one
+            2. the OSDs of the new (virtual) matching are randomly assigned to the actual (current OSDs), i.e.,
+            no matter which OSD has which folders.
+        while any algorithm (solving/approximating that kind of problem) could be used for the first step,
+        we here only implement the LPT algorithm, as it is a pretty good approximation with extremely good running time.
+        :return:
+        """
+        virtual_distribution = copy.deepcopy(self)
+        virtual_distribution.rebalance_lpt(rebalance_factor=0)
+
+        movements = {}
+
+        for virtual_osd in virtual_distribution.OSDs.values():
+            for virtual_folder in virtual_osd.folders.keys():
+                if not self.OSDs[virtual_osd.uuid].contains_folder(virtual_folder):
+                    movements[virtual_folder] = (self.get_containing_osd(virtual_folder).uuid, virtual_osd.uuid)
+
+        for virtual_osd in virtual_distribution.OSDs.values():
+            self.replace_osd(virtual_osd)
+
+        return movements
+
     def get_suitable_osds(self, folder_size):
         """
         create a list of OSDs with at least folder_size free capacity.