|
From: Jacob S. <dev...@js...> - 2011-05-06 12:02:17
|
Hi, Björn, thank you for your sugestions, I tried adding the values to sysctl, but it sadly did not change the problem. Most similar problems I found on "google" seemed to be related to a faulty bios, kernel or drivers. I have changed both machine and nic's, only kernel left on that track. I also tried disable IPv6 but it did not change anything. Kind Regards Jacob On Fri, 6 May 2011 08:18:21 +0200, Rudner, Björn <br...@ba...> wrote: > Hi Jacob, > > I remember that I also had this problem a while ago. > > > > I used heartbeat and apache on two servers with virtual-IP and > loadbalancing. I also encountered the problem with the ksoftirq on 100%. > > > > I did not exactly remember what solved the problem but I have done some > changes: > > 1. I switched from heartbeat to keepalived. > > 2. I have done some tweaks in the sysctl. I do not exactly know > which of them solved the problem but you might try it. > > Here are the settings from my sysctl.conf that could be involved: > > > > kernel.panic=3 > > #net.ipv4.netfilter.ip_conntrack_max=65536 > > net.ipv4.tcp_ecn=1 > > net.ipv4.tcp_fin_timeout=30 > > net.ipv4.tcp_keepalive_time=120 > > net.ipv4.tcp_timestamps=0 > > net.ipv4.netfilter.ip_conntrack_tcp_timeout_established=300 > > net.ipv4.netfilter.ip_conntrack_udp_timeout=60 > > net.ipv4.netfilter.ip_conntrack_udp_timeout_stream=180 > > > > # Set conntrack-stuff higher > > net.netfilter.nf_conntrack_max=3000000 > > > > Cheers, > > > > Björn Rudner > > > > > > > > > > Von: Devil-Linux [mailto:dev...@js...] > Gesendet: Donnerstag, 5. Mai 2011 16:43 > An: dev...@li... > Betreff: [Devil-Linux-discuss] ksoftirqd 100% > > > > Hi all, > > We just upgraded to latest software on one router/VPN/firewall. > > I had to change to ipsec-tools instead of *swan, but else all seemed fine > from the 1.2.13 we had before. > > Anyways, as soon as we put trafik through (VPN) it it become totaly > unresponcive, and top always hangs having either close to 100%si or close > to 100% on ksoftirqd. > > The system is stale for a minute or so and then clears upp again, probably > because the services runing trafic through it timeout. > >>From watching /proc/interrupts it seem to be the network cards generating >>alot of interupts. > > We have tried changing networkcards from 3com to intel. We even bought a > new machine, to change everything but still get the same problem. > > The new machine works better though as the IRQ only hog one CPU/core att a > time so the system still answers. > > My next step is to try and recompile Devil-Linux with noapic and try that, > but before I go ahead I thought I might ask here and see if you guys know > anything else I could try. It might be just some silly mistake. > > Kind Regards > > Jacob Sandin |