Hi Tim, On 16 April 2018 12:48 Phil Edworthy wrote: > On 16 April 2018 12:05 Tim Sander wrote: > > Am Montag, 16. April 2018, 10:21:04 CEST schrieb Phil Edworthy: > > > On 29 March 2018 08:30, Phil Edworthy wrote: > > > > On 28 March 2018 16:32, Clark Williams wrote: > > > > > On Wed, 28 Mar 2018 16:56:27 +0200 John Ogness wrote: > > > > > > On 2018-03-28, Phil Edworthy wrote: > > > > > > >> > I found that cyclictest results vary from one run to another. > > > > > > >> > > > > > > > >> > [...] > > > > > > >> > > > > > > > >> > Is it common knowledge that cyclictest results vary so much > > > > > > >> > from one run to another? Any ideas how to mitigate this? > > > > > > [snip] > > > > > > > > > > cyclictest -m -n -Sp99 -i200 -h300 -M -D 10h > > > > > > [snip] > > > > > > > Ok, I have changed the pri to 98, no difference in the results that > > > > I can see. > > > > > > > > I did some overnight tests with 100 runs of cyclictest running for 1 > > > > minute. Stats below were calculated using stats package from > > > > http://web.cs.wpi.edu/~claypool/misc/stats/stats.html > > > > > > > > 1. Interval fixed to 400us, not using --secalign > > > > Min: 20 Avg: 37 Max: 187 (avg of 100xMax is 134) > > > > > > > > 3. Interval increases from 400 to 499, not using --secalign > > > > Min: 20 Avg: 37 Max: 211 (avg of 100xMax is 157) > > > > > > I've got a bit further with this. > > > The max latency seen by cyclictest depends on the type of network load > > > used during the test (duh!). I use a host PC to ping flood the board > > > under test, however on my board (a low-end ARM board), the *size* of > > > the ping affects the measured latencies. > > > Using a ping size of 65506, cyclictest reports significantly higher > > > average and max latency than if I use the default ping size of 56 bytes. > > > > > > In addition to the ping size affecting the results, the cyclictest > > > interval also affects the latency seen. By running cyclictest for > > > short durations with the interval sweeping from 400us to 500us, I get > > > significantly higher max latency than just using an interval of 400us for > > many hours. > > > > > > So, is ping flood a good networking load for cyclictest? > > > Any suggestions for non-network related loads that can be used whilst > > > running cyclictest? Then I can determine if it is the GMAC > hardware/driver. > > I would recommend iperf3 for running load tests. You can't be sure how > > much of these ICMP Packages is handled in hardware. > Right, though in this case I am running on an embedded device and the MAC > is built into the SoC. Having checked in the datasheet, the MAC can only do > checksum hardware acceleration for ICMP. btw, I tried iperf3 but after a few runs for reasonable amount of time, it doesn't trigger max latencies anywhere near what I have been getting with a 64KB ping flood. Thanks anyway Phil -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
