Re: loopback behaviour under high load ...

[Davide Libenzi]

On Wed, 2 Apr 2003, Martin Josefsson wrote:

> On Wed, 2003-04-02 at 18:23, Davide Libenzi wrote:
> > On Tue, 1 Apr 2003, Davide Libenzi wrote:
> >
> > > And netstat shows Recv-Q=0 for the server socket, and Send-Q=N for the
> > > client socket. This has been tested on 2.5.66 vanilla.
> >
> > An update on this. Kernel 2.4.20+epoll does not have this problem.
>
> I see something similar with 2.5.64-bk10 (seen with earlier 2.5 kernels
> as well).
>
> I use distcc on my machine which is running 2.5.64-bk10 and using two
> remote servers running 2.4.20 to do the compiling.
>
> Sometimes I see a ~120s stall in the middle of a connection with
> Recv-Q=0 on the server (2.4.20) and Send-Q=N on the client (2.5.64-bk10)
> (it's the client that's sending data to the server)
>
> x.x.x.x = client
> y.y.y.y = server
>
> As you can see there's a long delay between 16:02:09 and 16:04:24
>
> There's some packetloss here (probably a congested interface in the
> router on the way). The dump was made on the client.

Hmm, we don't see any dropped packet at interface ( loopback ) level here:

$ cat /proc/net/softnet_stat
002a0c14 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000

$ ifconfig
lo        Link encap:Local Loopback
          inet addr:127.0.0.1  Mask:255.0.0.0
          UP LOOPBACK RUNNING  MTU:16436  Metric:1
          RX packets:1349417 errors:0 dropped:0 overruns:0 frame:0
          TX packets:1349417 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0

- Davide