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.
As we will see, contrary to the previous runs, the results now differ. The reason for this is that we started to support offloading on IPv6.
x hydra1/test_netperf_tcp6_default.stat
+ hydra1/test_netperf_tcp6_nooffload.stat
+--------------------------------------------------------------------------+
| + |
| + x |
| + x |
| + x |
| + + x |
| + + x |
| +++ x |
| ++++ x |
| ++++ x x x x |
|+++++ xx x xx x x |
|++++++ xxxxxxxxxx x x|
| |___A_M_| |
| |AM |
+--------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 30 7779.65 9129.07 8325.38 8247.0433 284.03515
+ 30 4012.84 4329.77 4193.7 4172.7127 82.614191
Difference at 95.0% confidence
-4074.33 +/- 108.121
-49.4035% +/- 1.31103%
(Student's t, pooled s = 209.166)
Given we have seen in the previous section, that IPv6 offload support makes a huge difference, we have added the 4th case to the following diagrams.
As expected IPv4 and IPv6 without offload support remain noticably different as already seen in the initial run.
Contrary to that IPv6 with offload support is almost closing the gap to IPv4 now.
To get a better view on how far IPv6 is behind IPv4 with and without offloading, we also zoom comparing just these two cases.
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
+--------------------------------------------------------------------------+
| % |
| % |
| % |
| % |
| % |
| *% |
| *% |
| *% |
| x x + *% |
| x x ++ *% |
| xxx ++ *%% |
| xxxx +++ *%% |
| xxxx +++++ * * *%%*% |
|xxxxx +++++++ ** * **%*%%O%% |
|xxxxxx +++++++++ + *******OOO%O%% *|
| |AM |
| |__A_| |
| |___A_M_| |
| |MA| |
+--------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 30 4012.84 4329.77 4193.7 4172.7127 82.614191
+ 30 5295.53 5945.17 5728.22 5667.635 168.46945
Difference at 95.0% confidence
1494.92 +/- 68.5833
35.8262% +/- 1.64361%
(Student's t, pooled s = 132.678)
* 30 7779.65 9129.07 8325.38 8247.0433 284.03515
Difference at 95.0% confidence
4074.33 +/- 108.121
97.6423% +/- 2.59114%
(Student's t, pooled s = 209.166)
% 30 8300.31 8710.39 8448.19 8476.1713 104.14566
Difference at 95.0% confidence
4303.46 +/- 48.5891
103.133% +/- 1.16445%
(Student's t, pooled s = 93.9985)