2012-03-21 (2011-12-15) p4 @203041 TCP loopback

Test runs

netperf-2.5 TCP_STREAM test was run over 3 reboot iterations, with each iteration having 1 warm up run and 10 recorded netperf runs.

IPv6/TCP with and without offloading turned on

As we can see, the median of boths IPv6 runs is close to equal. However the range of the data and the interquartile range (IQR) on loopback is extremly high.

IPv6/TCP numbers with and without offloading on IPv6/TCP numbers with and without offloading on
x hydra1/test_netperf_tcp6_default.stat
+ hydra1/test_netperf_tcp6_nooffload.stat
+--------------------------------------------------------------------------+
|                                                             +            |
|         x                 x     +                           +            |
|       + xx         x      x   + +                           +            |
|x    x + **+x   +++xx+    +*x+*+**xx +   +  x xx x*   x + *+ ++   x  x   x|
|            |_________________M_A___________________|                     |
|                |_______________M_A__________________|                    |
+--------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x  30       4105.23       6099.56       4914.33     4971.6027     551.08148
+  30       4295.29       5789.85       4979.33     5040.9523     508.45581
No difference proven at 95.0% confidence
		

IPv4 and IPv6 TCP with and without offloading

Given IPv6 offload support has not yet been implemented, and given we have seen in the previous section that there was no real difference, we will only compare IPv4 with and without offloading, to IPv6 default.

As expected IPv4 and IPv6 without offload differ. Once again IPv4 without offloading turned on is doing better than IPv6.

IPv4 TCP w/ and w/o offloading, IPv6 TCP default (no offloading) IPv4 TCP w/o offloading, IPv6 TCP default
x hydra1/test_netperf_tcp6_nooffload.stat
+ hydra1/test_netperf_tcp4_nooffload.stat
* hydra1/test_netperf_tcp6_default.stat
% hydra1/test_netperf_tcp4_default.stat
+--------------------------------------------------------------------------+
|     *       * xx          +  x++         %  %                            |
|    **  x *  **xx*    *+*+ + +x++     + %%% %%%   %      %                |
|*  **** x**  ****** x ***+ ***x+*+*+* +%%%@ %@@% %%% %%%%% % %     %     %|
|        |_______MA________|                                               |
|                        |______A_____|                                    |
|      |_______MA_________|                                                |
|                                         |_____M_A________|               |
+--------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x  30       4295.29       5789.85       4979.33     5040.9523     508.45581
+  30       5106.56       6652.14       5853.92      5817.592     366.75431
Difference at 95.0% confidence
	776.64 +/- 229.15
	15.4066% +/- 4.54577%
	(Student's t, pooled s = 443.304)
*  30       4105.23       6099.56       4914.33     4971.6027     551.08148
No difference proven at 95.0% confidence
%  30       6265.38       8191.13       6726.55      6875.803     476.59095
Difference at 95.0% confidence
	1834.85 +/- 254.725
	36.3989% +/- 5.05312%
	(Student's t, pooled s = 492.781)