On 09/18/2013 10:52 AM, Bob Copeland wrote: > On Tue, Sep 17, 2013 at 04:33:34PM -0700, Adrian Chadd wrote: >> ... I think you should start doing some more in-depth system profiling to >> see where that CPU is actually being used. Saying "ksoftirqd" is only a >> little helpful. >> >> See if it's handling way too much interrupts, or whether there's lots of >> card error conditions, or the TX/RX queues are being too frequently >> checkeed, or ... >>> What I have seen is that with madwifi driver the cpu is 90-100% cpu >>> usage with ksoftirqd with mac80211 about 2% ... >>> Could this be a kernel interrupt issue? Are there several policies which >>> changed with kernel? > Indeed I didn't understand at all the stats given above. I took it to mean > that madwifi is using much more CPU in the kernel than ath5k, which would be > an anti-feature unless ath5k is sleeping too much. At any rate - results of Yes this is what I have seen. If using madwifi driver, iperf is using 10% cpu and almost 80-90% cpu BUT throughput is nice: 13-15Mbps@24M bit rate fix where with ath5k iperf is using 3% cpu and interrupts nothing and throughput is only 5-7Mbps. Hostapd is used with wpa2-psk. > "perf record [do something wireless]" + "perf report" for both scenarios > could shed some light. > What do you want to know?
On machine where hostap is running: while true; do iperf -c <ip_of_sta>;done on this machine ksoftirqd/0 use no cpu with top, iperf 1% cpu if at5hk driver is used and 80-90% cpu for ksoftirqd/0 and about 10% for iperf if madwifi is used. On machine were wpa_supplicant is running iperf -s top shows about 2-5% cpu of iperf and nothing about ksoftirqd/0 nevertheless if ath5k or madwifi is used. and visa vers on machine where hostap is running: iperf -s top shows about 2-5% cpu of iperf and nothing about ksoftirqd/0 nevertheless if ath5k or madwifi is used. on machine were wpa_supplicant is running while true; do iperf -c <ip_of_master>done on this machine ksoftirqd/0 use no cpu with top, iperf 1% cpu if at5hk driver is used and 80-90% cpu for ksoftirqd/0 and about 10% for iperf if madwifi is used. Nevertheles if machine were wpa_supplicant is running is using madwifi or ath5k driver throughput is 13-15Mbps IF machine where hostap is running use madwifi driver If machine where hostap is running use ath5k driver throughput is 5-7Mbps nevertheles if machine were wpa_supplicant is running is using madwifi or ath5k driver. Other "problem" what I see is that the signal is normaly -60 and drops out short time to -78 with ath5k with madwifi not ... Is that what you want to known? BTW: I'm now compiling the kernel with debug support for ath5k drivers _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
