netperf-2.5 UDP_STREAM test was run over
3 reboot iterations, with each iteration having
1 warm up run and 10 recorded netperf runs.
netperf needed a patch to be able to run this test
on IPv6, which by now is part of the port and the
upstream trunk revision.
On IPv6 UDP checksums are mandatory, so we did not turn them off for comparisons.
As we expect the IPv4 tests with checksumming turned off perform best. However IPv6 UDP is behind IPv4 UDP with checksumming turned on and zooming in gives us a clearer view.
x hydra1/test_netperf_udp4_default_nocksum.stat + hydra1/test_netperf_udp4_nooffload_nocksum.stat * hydra1/test_netperf_udp4_default.stat % hydra1/test_netperf_udp4_nooffload.stat # hydra1/test_netperf_udp6_default.stat @ hydra1/test_netperf_udp6_nooffload.stat +--------------------------------------------------------------------------+ | O | | O | | @O O* | | @O OO | | OO@ %OO | | # OO@ % %OO *x | | # OO@ % %OO +** | | @ ##OO@ %*OOO x+ x**** | |#@###OO@ %*OOO x+x+ *****+ +| |#@OOOOO@ % % %*%*OOOO +x+*+x******++| | |_AM_| | | |__AM__| | | |A| | | |__AM| | | |AM| | | |_AM| | +--------------------------------------------------------------------------+ N Min Max Median Avg Stddev x 30 5368.02 5558.67 5508.52 5490.151 54.008218 + 30 5333.76 5607.16 5502.8 5493.3683 66.594082 No difference proven at 95.0% confidence * 30 5107.9 5219.41 5186.66 5181.557 25.270552 Difference at 95.0% confidence -308.594 +/- 21.7948 -5.62087% +/- 0.39698% (Student's t, pooled s = 42.1633) % 30 4999.33 5221.67 5182.04 5159.3187 51.909327 Difference at 95.0% confidence -330.832 +/- 27.3805 -6.02592% +/- 0.498721% (Student's t, pooled s = 52.9692) # 30 4128.63 4259.98 4222.2 4217.663 37.613551 Difference at 95.0% confidence -1272.49 +/- 24.0564 -23.1777% +/- 0.438174% (Student's t, pooled s = 46.5385) @ 30 4151.75 4271.11 4247.59 4232.549 35.397842 Difference at 95.0% confidence -1257.6 +/- 23.6029 -22.9065% +/- 0.429914% (Student's t, pooled s = 45.6612)