[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <200808271509.18529.denys@visp.net.lb>
Date: Wed, 27 Aug 2008 15:09:17 +0300
From: Denys Fedoryshchenko <denys@...p.net.lb>
To: Evgeniy Polyakov <johnpol@....mipt.ru>
Cc: Eric Dumazet <dada1@...mosbay.com>, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: loaded router, excessive getnstimeofday in oprofile
On Tuesday 26 August 2008, Evgeniy Polyakov wrote:
> On Tue, Aug 26, 2008 at 10:44:56PM +0200, Eric Dumazet (dada1@...mosbay.com)
wrote:
> > >Do you have any packet sockets in this system? Like running dhcp daemon?
No, nothing at all.
> >
> > Another way to see this problem can be to start a sniffer on the machine,
> > even with a restrictive pcap filter, to check if performance change or
> > not. (It should decrease)
Yes, when i run tcpdump even without promisc at peak time, machine will be
almost dead. Transit traffic will be 100ms+. I know that it is timestamping
packets. Same almost for any libpcap app.
>
> Or just check /proc/net/packet iirc.
> Anyway, having at least one packet socket ends up with timestamping of
> each packet, so you will get fair load of getnstimeofday() in that case.
There is very short list of tasks. Attached.
/proc/net/packet clean, nothing there.
>
> > For example, I believe that running "ping" could have the same effect
> > (increasing netstamp_needed variable : every incoming packet has to be
> > timestamped)
Even answering icmp timestamp request will take resources.
> >
> > So beware of pings, traceroute and other networking tools...
When i am measuring performance - they are all off.
>
> Yup, this innocent toys can end up with this such behaviour on modern
> highly loaded machines.
View attachment "tasks.txt" of type "text/plain" (1170 bytes)
Powered by blists - more mailing lists