traffic_replay: Prevent users having 1000+ memberOf links

When adding 10,000 users, one user would end up in over 1000 groups.
With 100,000 users, it would be more like 10,000 groups. While it makes
sense to have groups with large numbers of users, having a single user
in 1000s of groups is probably less realistic.

This patch changes the shape of the Pareto distribution that we use to
assign users to groups. The aim is to cap users at belonging to at most
~500 groups. Increasing the shape of the Pareto distribution pushes the
user assignments so they're closer to the average, and the tail (with
users in lots of groups) is not so large).

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

diff --git a/python/samba/emulate/traffic.py b/python/samba/emulate/traffic.py
index 16672286cd89dde5ab2cbc781236af659820a86a..aabf6ed0a4225bd4e2581b9f8c404a444700ff62 100644 (file)
--- a/python/samba/emulate/traffic.py
+++ b/python/samba/emulate/traffic.py
@@ -1818,7 +1818,7 @@ class GroupAssignments(object):
                  users_added, group_memberships):
 
         self.generate_group_distribution(number_of_groups)
-        self.generate_user_distribution(number_of_users)
+        self.generate_user_distribution(number_of_users, group_memberships)
         self.assignments = self.assign_groups(number_of_groups,
                                               groups_added,
                                               number_of_users,
@@ -1839,15 +1839,27 @@ class GroupAssignments(object):
             dist.append(cumulative / total)
         return dist
 
-    def generate_user_distribution(self, n):
+    def generate_user_distribution(self, num_users, num_memberships):
         """Probability distribution of a user belonging to a group.
         """
         # Assign a weighted probability to each user. Use the Pareto
         # Distribution so that some users are in a lot of groups, and the
-        # bulk of users are in only a few groups
+        # bulk of users are in only a few groups. If we're assigning a large
+        # number of group memberships, use a higher shape. This means slightly
+        # fewer outlying users that are in large numbers of groups. The aim is
+        # to have no users belonging to more than ~500 groups.
+        if num_memberships > 5000000:
+            shape = 3.0
+        elif num_memberships > 2000000:
+            shape = 2.5
+        elif num_memberships > 300000:
+            shape = 2.25
+        else:
+            shape = 1.75
+
         weights = []
-        for x in range(1, n + 1):
-            p = random.paretovariate(1.0)
+        for x in range(1, num_users + 1):
+            p = random.paretovariate(shape)
             weights.append(p)
 
         # convert the weights to a cumulative distribution between 0.0 and 1.0