Large IP packets do not flow

[Ilya Shakhat]

Even if private interface is configured with MTU 9000 IP packets larger than 1500 do not reach destination.

Steps to reproduce:
  1. Provision 2 instances in the same tenant network (L2 domain) and on different compute nodes
  2. Run iperf between nodes:
       ubuntu@shaker-apzokt-master-0:~$ iperf -c 10.0.0.5 -t 10 -M 1460 -- succeed, shows > 9G
       ubuntu@shaker-apzokt-master-0:~$ iperf -c 10.0.0.5 -t 10 -M 1470 -- hangs

 Deployment:
  3 controllers, 3 computes, VLAN, Ubuntu
Network configuration:
  eth0 - public
  eth1 - <none>
  eth2 - admin, storage, mgmt
  eth3 - private, MTU 9000
On all interfaces 'disable offloading' switched off.

VERSION:
  feature_groups:
    - mirantis
  production: "docker"
  release: "6.1"
  openstack_version: "2014.2.2-6.1"
  api: "1.0"
  build_number: "521"
  build_id: "2015-06-08_06-13-27"
  nailgun_sha: "4340d55c19029394cd5610b0e0f56d6cb8cb661b"
  python-fuelclient_sha: "4fc55db0265bbf39c369df398b9dc7d6469ba13b"
  astute_sha: "7766818f079881e2dbeedb34e1f67e517ed7d479"
  fuel-library_sha: "f43c2ae1af3b493ee0e7810eab7bb7b50c986c7d"
  fuel-ostf_sha: "7c938648a246e0311d05e2372ff43ef1eb2e2761"
  fuelmain_sha: "bcc909ffc5dd5156ba54cae348b6a07c1b607b24"

[Andrey Maximov (maximov)]

our preliminary analysis shows that the possible reason - jumbo frames are not allowed on routers. moving to scalelab

[Dina Belova (dbelova)]

This bug has been fixed in our lab configuration.