On Sep 28, 2014, at 11:45 , Sebastian Moeller <moeller0@gmx.de> wrote:
[…]
Now, I have performed a few netperf-wrapper tests recently and it seems there is something in my system that screws up latency under load roughly every minute (see attached image). Until recemty I believed these to be caused by my MacBooks wireless access, but the attached image is from a test win se00 (also no soft-irq spikes visible on top running on cerowrt). I begin to wonder whether these might be caused by ohcpd’s repeated search for a useable IPv6 link, but I digress.
This is not going to be too interesting, but for some closure I want to document the information about this issue (especially since cerowrt is completely innocent, and I want to avoid spoiling its reputation with another device’s issues)
So I figured that one out; it was the 2.4GHz wifi on my ISP’s modem-router (speed port w723v type A, avoid if possible); which was a) running without associated stations as an emergency fallback in case I screwed up cerowrt, and b) was actually wedging itself hard after 1 to 48 hours of operation in that it did not allow any station to connect. And in the wedged state it completely wiped out this modem-router’s wan throughput. Since this router was shared between my wired and wireless tests (after all this is the device talking DSL ;) ) it affected all my tests. Since I disabled that devices WLAN completely, the latency spikes are gone (actually now I have periodic spikes, but these are of much smaller magnitude and much less urgent to solve).\\\
Best Regards
Sebastian