OpenWrt Forum Archive

Topic: ksoftirqd eats up all the CPU

The content of this topic has been archived on 18 Apr 2018. There are no obvious gaps in this topic, but there may still be some posts missing at the end.

Hi,

with fast BT-Downloads (1,2MB/s), ksoftirqd/0 eats up all the CPU on Kamikaze 5461. Can somebody confirm this? I did the Connection-Settings from the Azureus-Wiki

download goes up to 1,2mb/s for a few seconds and then down for a long time.

loadavg: 1.93 1.31 0.58 2/46 1327 on the router. ssh lags very much


Hardware is a WGT634U with PPPoE and bridged LAN/WLAN.

This is normal behaviour for a heavily loaded CPU.
The text below is taken from http://www.opensourcemanuals.org/manual/ksoftirqd/

ksoftirqd is a per-cpu kernel thread that runs when the machine is under heavy soft-interrupt load.
Soft interrupts are normally serviced on return from a hard interrupt, but it?s possible for soft interrupts to be triggered more quickly than they can be serviced.
If a soft interrupt is triggered for a second time while soft interrupts are being handled, the ksoftirq daemon is triggered to handle the soft interrupts in process context.
If ksoftirqd is taking more than a tiny percentage of CPU time, this indicates the machine is under heavy soft interrupt load.

The 200mhz mipsel processors in Broadcom routers (which the wgt is) simply aren't that powerful.  They tend to top out at 15-20Mbps when the processor has to route the traffic.  Throw into the mix that BitTorrent is fairly demanding when transferring from many sources at high speed and it looks like you've hit the limit of your router.

alright its fine with the defaults from launchmany-curses now. max global conns = 80 per default i think.

loadavg is 0.24 0.12 0.04 2/29 803

some hints for deeper researches:
echo "1024" >  /sys/module/ip_conntrack/parameters/hashsize
echo "4096" > /proc/sys/net/ipv4/netfilter/ip_conntrack_buckets
smoothwall-thread

The discussion might have continued from here.