I don't think I'm up to the challenge of an ASCII art representation of
our cluster. How's this instead?
We run on a cluster of 1U and 2U servers. Each server is connected to
the same network by either multiple 1GigE or 10GigE links and is
configured with a bonded/bridged interface. Every server runs some
number of containers, based on resources available and the configuration
parameters we select when installing the cluster. Almost all network
traffic is between the various containers, both within the same host as
well as between containers running on different hosts. There is minimal
communication in our software between containers and the servers
themselves. We have some protocols based on UDP and a few based on TCP.
Pretty standard stuff ultimately.
We run this identical configuration with VMs (KVM) without issues. We're
moving to containers since our benchmarks indicate we'd gain some in
performance.
On 09/07/2015 10:35 PM, Guido Jäkel wrote:
Dear Peter,
may you paint a small ASCII art of your desired network setup and the network
configuration? I see no reason that "LXC", i.e. the veth connecting two
differently namespaced IP stacks is responsible for any problems.
BTW: I'm using a complex network setup with no problems, too. A real NIC is
connected to a selected trunk of VLANs; they are decoded and connected to a
couple of bridges and the containers are attached to one or more of it as
needed for the business case.
greetings
Guido
On 07.09.2015 20:49, Peter Steele wrote:
We're having issues with networking connections in our containers when the host
is configured with bonded interfaces. When we configure these same servers to
run with VMs, everything works fine, but when we swap out the VMs for
equivalently configured containers, we get all kinds of network connection
failures between containers. The network issues go away when we reduce the bond
to have a single member. More specifically, we configure our server with a host
bridge (br0), where the host bridge links to the host's bonded interface
(bond0). We've been using this model for years with VMs. Something isn't
working quite right with containers though.
Is there something we need to set in our containers' config files when the host
is setup with bonding?
_______________________________________________
lxc-users mailing list
lxc-users@lists.linuxcontainers.org
http://lists.linuxcontainers.org/listinfo/lxc-users
_______________________________________________
lxc-users mailing list
lxc-users@lists.linuxcontainers.org
http://lists.linuxcontainers.org/listinfo/lxc-users
_______________________________________________
lxc-users mailing list
lxc-users@lists.linuxcontainers.org
http://lists.linuxcontainers.org/listinfo/lxc-users