Retry-loop for task set generation

Task set generation put in a retry loop. if a task set fails
to be partitionable under any partition scheme, the entire task
set is retried.

1 files changed, 67 insertions, 43 deletions

diff --git a/gen/generator.py b/gen/generator.py
index 2d0fbef..fcb21eb 100644
--- a/gen/generator.py
+++ b/gen/generator.py
@@ -137,6 +137,8 @@ class Generator(object):
         # make pgmrt arguments using graphs and tasks.
         sched_file = self.out_dir + "/" + FILES['sched_file']
 
+        # task set is in microseconds. we must convert to milliseconds
+
         graph_desc_arg = []
         rates_arg = []
         exec_arg = []
@@ -153,12 +155,17 @@ class Generator(object):
             split_arg_t = []
 
             for n in g.nodes:
+                # task set is in microseconds. we must convert to milliseconds
+
                 cluster_arg_t.append('node_' + str(n.id) + ':' + str(n.task.partition))
-                exec_arg_t.append('node_' + str(n.id) + ':' + str(n.task.cost))
-                split_arg_t.append('node_' + str(n.id) + ':' + str(n.task.split))
+                cost_str = format(n.task.cost/1000.0, '.4f').rstrip('0').rstrip('.')
+                exec_arg_t.append('node_' + str(n.id) + ':' + cost_str)
+                if n.task.split != 1:
+                    split_arg_t.append('node_' + str(n.id) + ':' + str(n.task.split))
                 if n.isSrc == True:
                     # assume that x=1
-                    rates_arg_t.append('node_' + str(n.id) + ':1:' + str(n.task.period))
+                    period_str = format(n.task.period/1000.0, '.4f').rstrip('0').rstrip('.')
+                    rates_arg_t.append('node_' + str(n.id) + ':1:' + period_str)
                     # get cluster size
                     clustersz_arg_t = str(pgm_params['cpus'] / pgm_params['nr_clusters'])
                 if len(g.nodes) == 1:
@@ -167,7 +174,8 @@ class Generator(object):
                     graph_desc_arg_t.append('node_' + str(n.id) + ':node_' + str(succ.id))
                 # wss parameter
                 for e in n.outEdges:
-                    wss_arg_t.append('node_' + str(n.id) + ':node_' + str(e.s.id) + ':' + str(e.wss))
+                    wss_kb_str = format(e.wss, '.4f').rstrip('0').rstrip('.')
+                    wss_arg_t.append('node_' + str(n.id) + ':node_' + str(e.s.id) + ':' + wss_kb_str)
 
             # combine arguments to a comma-separated string
             cluster_arg_t = ','.join(cluster_arg_t)
@@ -192,7 +200,8 @@ class Generator(object):
             pgm_args_t = '';
             pgm_args_t += '--wait --cluster ' + cluster_arg[i] + ' --clusterSize ' + clustersz_arg[i] + ' --name graph_' + str(i)
             pgm_args_t += ' --graph ' + graph_desc_arg[i] + ' --rates ' + rates_arg[i] + ' --execution ' + exec_arg[i]
-            pgm_args_t += ' --split ' + split_arg[i]
+            if len(split_arg[i]) != 0:
+                pgm_args_t += ' --split ' + split_arg[i]
             if len(wss_arg[i]) != 0:
                 pgm_args_t += ' --wss ' + wss_arg[i]
 
@@ -366,44 +375,59 @@ class Generator(object):
                 dp.nr_sink = graph.uniform(opts.nr_sink, opts.nr_sink)
                 dp.wss = NAMED_EDGE_WSS[dp.wss]
                 
-                # Generate a task set
-                ts, graphs, subts = self._create_tasks(dp)
-                dp.tasks = len(ts)
-
-                #for dp in PgmDesignPointGenerator(exp):
-                for pm in partition_method:
-                    dp.partitions = pm
-                    # Create directory name from relevant parameters
-                    dir_leaf  = "sched=%s_%s" % (self.scheduler, col_map.encode(dp))
-                    dir_leaf  = dir_leaf.strip('_') # If there are none
-                    dir_leaf += ("_trial=%s" % trial) if trials > 1 else ""
-
-                    dir_path  = "%s/%s" % (out_dir, dir_leaf.strip('_'))
-
-                    if os.path.exists(dir_path):
-                        if force:
-                            sh.rmtree(dir_path)
-                        else:
-                            print("Skipping existing experiment: '%s'" % dir_path)
-                            continue
-
-                    os.mkdir(dir_path)
-
-                    if trials > 1:
-                        dp[PARAMS['trial']] = trial
-                    self.out_dir = dir_path
-
-                    dp.system = topology.Topology(machines[dp.host])
-
-                    # Write a sched.py and param.py for each partition method
-                    ret = self._create_exp(dp, ts, graphs, subts)
-                    if not ret:
-                        print("Bin-packing fails for " + dir_leaf)
-                        os.rmdir(dir_path)
-
-                    del(self.out_dir)
-                    if PARAMS['trial'] in dp:
-                        del dp[PARAMS['trial']]
+                last_failed = ''
+                tries = 0
+                success = False
+                while tries < 100 and not success:
+                    created_dirs = []
+                    tries += 1
+
+                    # Generate a task set
+                    ts, graphs, subts = self._create_tasks(dp)
+                    dp.tasks = len(ts)
+
+                    try:
+                        for pm in partition_method:
+                            dp.partitions = pm
+                            # Create directory name from relevant parameters
+                            dir_leaf  = "sched=%s_%s" % (self.scheduler, col_map.encode(dp))
+                            dir_leaf  = dir_leaf.strip('_') # If there are none
+                            dir_leaf += ("_trial=%s" % trial) if trials > 1 else ""
+
+                            dir_path  = "%s/%s" % (out_dir, dir_leaf.strip('_'))
+
+                            if os.path.exists(dir_path):
+                                if force:
+                                    sh.rmtree(dir_path)
+                                else:
+                                    print("Skipping existing experiment: '%s'" % dir_path)
+                                    continue
+
+                            os.mkdir(dir_path)
+                            created_dirs.append(dir_path)
+
+                            if trials > 1:
+                                dp[PARAMS['trial']] = trial
+                            self.out_dir = dir_path
+
+                            dp.system = topology.Topology(machines[dp.host])
+
+                            # Write a sched.py and param.py for each partition method
+                            ret = self._create_exp(dp, ts, graphs, subts)
+                            if not ret:
+                                print("Bin-packing fails for " + dir_leaf)
+                                last_failed = dir_leaf
+                                raise Exception("Failed to partition.")
+                            del(self.out_dir)
+                            if PARAMS['trial'] in dp:
+                                del dp[PARAMS['trial']]
+                        success = True
+                    except Exception, e:
+                        for d in created_dirs:
+                            sh.rmtree(d)
+                if not success:
+                    print("Failed to generate experiments. Last failed: %s" % last_failed)
+
 
     def print_help(self):
         display_options = [o for o in self.options if not o.hidden]