________________________________
From: Lenny [mailto:[email protected]]
Sent: July 30, 2009 3:28 AM
On Wed, Jul 29, 2009 at 11:27 PM, Evgeny Yurchenko
<[email protected]> wrote:
From: Lenny [mailto:[email protected]]
Sent: July 29, 2009 3:38 PM
Evgeny Yurchenko wrote:
I would try to swap cables and
interfaces in config and see errors. Do they go to em0? stay on em1?
It's pointless trying to fix tcp/ip without eliminating problem on
media.
Eugene.
That's all understandable when speaking of
errors and packet loss, but would it really cause the CPU hit 100% at
50kpps?
both em0 and em1?
By the way, it worked for 3 weeks with regular
load (about 10kpps) and the CPU was around 20%(each) and there were no
errors.
Lenny.
10kpps - 20% CPU
50kpps - 100% CPU
looks like we have some logic here. I've looked
at my graphs - there is no relation between cpu load and pps. Do you
have this relation? Not talking about your extreme case 50kpps,
generally - when load fluctuates let's say 10 to 15kpps, does you cpu
load also goes higher/lower?
I experienced 100% CPU only in two cases:
1) Multicast went from LAN to WAN and caused
storm (in carped setup)
2) There is known bug with slbd.
Eugene
I'm attaching links to the RRD graphs from the same period.
http://img119.imageshack.us/img119/1573/statusrrdgraphimgcpu.png
http://img253.imageshack.us/img253/4677/statusrrdgraphimgpacket.png
http://img248.imageshack.us/img248/4779/statusrrdgraphimgtraffi.png
http://img75.imageshack.us/img75/168/statusrrdgraphimgqualit.png
Lenny.
Weird, I do not have any relation between cpu and
bandwidth/packets:
http://img43.imageshack.us/img43/4127/bandwidth.png
http://img78.imageshack.us/img78/8375/cpu.png
http://img78.imageshack.us/img78/5235/packets.png
Eugene
