[Users] CentOS 6: default inbound traffic limited for CT's

[Mattias Geniar]

Hi,

I'm running OpenVZ on a CentOS 6 x64 machine with a stable kernel
(2.6.32-042stab049.6) and have even tried the issue below on the
test-branch (2.6.32-042stab052.2).
The issue I'm experiencing is also reported by Fredericep on the forum
but never got any follow-up:
http://forum.openvz.org/index.php?t=tree&goto=44238&&srch=inbound#msg_44
238

It's the exact same problem: my hardware node runs perfectly fine, it
has both full in- and outgoing networking speed. I can consistently
(3hours+ as tested) download files at my full line speed.
Whenever I try to same in a container, I can get a quick burst of
network traffic for a few seconds (10MB/s+) and then fall back to
10-200Kb/s, it varies.

My first troubleshooting went to incoming traffic shaping for the
eth0/venet0 interface, but that's not the case:
  # tc -s qdisc ls dev eth0
  qdisc mq 0: root
   Sent 65350399 bytes 363366 pkt (dropped 0, overlimits 0 requeues 0)
   rate 0bit 0pps backlog 0b 0p requeues 0

  # tc qdisc del dev eth0 root
  RTNETLINK answers: No such file or directory

But this is a default install and it doesn't have any traffic shaping
rules active. Neither does it have iptables active. It's all still
running the default OpenVZ stack.

Second idea was a possible hit of TCPSNDBUF or TCPRCVBUF as the defaults
(1720320) are rather low. But even changing it to something idiotic like
9223372036854775807 didn't make a difference. The /proc/user_beancounter
also didn't report any failed packets.

When tcpdumping the stream, I don't see anything abnormal except that
it's just slow traffic. Nothing out of the ordinary at first glance.

I'm looking for any advice on how to troubleshoot this, as I believe
this may very well be a CentOS 6 kernel bug - but to prove that, I would
of course need to dive deeper which is where my train of thought kind of
ends.

I look forward to any reply/idea this list may give me.

Mattias Geniar