Index: Tools/ccbench/ccbench.py
===================================================================
--- Tools/ccbench/ccbench.py	(revision 79534)
+++ Tools/ccbench/ccbench.py	(working copy)
@@ -18,6 +18,7 @@
 import socket
 from optparse import OptionParser, SUPPRESS_HELP
 import platform
+import math
 
 # Compatibility
 try:
@@ -193,7 +194,7 @@
             t1 = t2
 
 
-def run_throughput_test(func, args, nthreads):
+def run_throughput_test(func, args, nthreads, do_yield):
     assert nthreads >= 1
 
     # Warm up
@@ -224,7 +225,7 @@
             while not started:
                 start_cond.wait()
         results.append(loop(start_time, THROUGHPUT_DURATION,
-                            end_event, do_yield=True))
+                            end_event, do_yield=do_yield))
 
     threads = []
     for i in range(nthreads):
@@ -245,8 +246,19 @@
         t.join()
 
     return results
+    
+def compute_balance(values):
+    #compute the normalized standard deviation of the results
+    if len(values) < 2:
+        return 0.0
+    avg = sum(values)/len(values)
+    var = sum((avg-e)**2 for e in values) / (len(values)-1)
+    stddev = math.sqrt(var)
+    #normalize stddev by average
+    balance = stddev / avg
+    return balance
 
-def run_throughput_tests(max_threads):
+def run_throughput_tests(max_threads, do_yield):
     for task in throughput_tasks:
         print(task.__doc__)
         print()
@@ -254,16 +266,17 @@
         nthreads = 1
         baseline_speed = None
         while nthreads <= max_threads:
-            results = run_throughput_test(func, args, nthreads)
+            results = run_throughput_test(func, args, nthreads, do_yield)
             # Taking the max duration rather than average gives pessimistic
             # results rather than optimistic.
             speed = sum(r[0] for r in results) / max(r[1] for r in results)
-            print("threads=%d: %d" % (nthreads, speed), end="")
+            balance = compute_balance([r[0] for r in results])
+            print("threads=%2d: %5d" % (nthreads, speed), end="")
             if baseline_speed is None:
-                print(" iterations/s.")
+                print(" iterations/s. balance")
                 baseline_speed = speed
             else:
-                print(" ( %d %%)" % (speed / baseline_speed * 100))
+                print(" (%3d%%)        %.4f" % (speed / baseline_speed * 100, balance))
             nthreads += 1
         print()
 
@@ -546,6 +559,9 @@
     parser.add_option("-n", "--num-threads",
                       action="store", type="int", dest="nthreads", default=4,
                       help="max number of threads in tests")
+    parser.add_option("-y", "--no-yield",
+                      action="store_false", dest="do_yield", default=True,
+                      help="don't force threads to yield")
 
     # Hidden option to run the pinging and bandwidth clients
     parser.add_option("", "--latclient",
@@ -593,7 +609,7 @@
     if options.throughput:
         print("--- Throughput ---")
         print()
-        run_throughput_tests(options.nthreads)
+        run_throughput_tests(options.nthreads, options.do_yield)
 
     if options.latency:
         print("--- Latency ---")