On Wed, Jun 06, 2012 at 09:18:12PM -0500, Dong Jin wrote:
> Jun 07 06:01:39|3|dpif_linux|ERR|Generic Netlink family
> 'ovs_datapath' does not exist. The Open vSwitch kernel module is
> probably not loaded.
> Jun 07 06:01:39|4|dpif|WARN|failed to enumerate system datapaths:
> No such file or
I combined the two commands as you suggested. However, the same error
message is still shown. I also tried to move those ovs-vsctl command with
--no-wait option before the ovs-vswitchd command, but the error remains.
Any suggestions? thanks.
On Thu, Jun 7, 2012 at 7:38 PM, Ethan Jackson wrote:
>
thanks, Ethan and Ben.
On Thu, Jun 7, 2012 at 7:29 PM, Ethan Jackson wrote:
> > So upgrade to 1.7.0 (see from another post discussing on performance,
> but it
> > is related to kernel mode) should not help, right?
>
> Correct, performance improvements were targeted at kernel datapaths.
> Some of
I think what you're seeing is a race condition.
> ovs-vsctl -- --may-exist add-br br0
Here, you've created a bridge of type system.
Time happens . . .
> ovs-vsctl set bridge br0 datapath_type=netdev
Here you change it to netdev.
You can fix the race condition by doing it all in one transactio
> So upgrade to 1.7.0 (see from another post discussing on performance, but it
> is related to kernel mode) should not help, right?
Correct, performance improvements were targeted at kernel datapaths.
Some of those improvements may have impacted the userspace datapaths a
but, but I doubt they had
On Thu, Jun 07, 2012 at 06:50:15PM -0500, Dong Jin wrote:
> Just wondering if improving the performance of user space datapath is on
> schedule or just something with low priority?
We don't have much motivation to do it ourselves at Nicira. I think
we'd accept patches to improve userspace perform
thanks a lot, Ethan. Below is the entire script I used to start the ovs
daemon, and netdev is used. Just wondering why I still see the warning and
error. Or can I safely ignore the warning and error?
ovsdb-server --remote=punix:/usr/local/var/run/openvswitch/db.sock
--remote=db:Open_vSwitch,manage
Thanks, Ethan. I am trying to run ovs in OpenVZ VM. Since OpenVZ uses
OS-level virtualization (all VM shares the same kernel) and unfortunately I
cannot load kernel module in VM, that's why I have to use the userspace
datapath.
So upgrade to 1.7.0 (see from another post discussing on performance,
Sounds like the datapath_type of the bridge you created is "system".
Since the kernel module is not loaded, it's failing to create the
bridge. You want "netdev" if you're planning to stick with userspace.
Ethan
On Wed, Jun 6, 2012 at 7:18 PM, Dong Jin wrote:
> 1. I am using the userspace mode o
> Thanks, Jesse. Could you explain more details on what features are not
> available in the user space datapath?
The main feature being performance. That being the issue you're
running into, I would suggest switching. Certain kinds of port types
aren't supported as well. Tunneling and what not.
I agree with your description below, however not being able to tell the status
of LACP is a significant issue for the sysadmin during network troubleshooting,
especially when the switch is missconfigured. It just seems odd to me you would
withhold a patch that set things straight on the OVS side
The patch doesn't actually fix a problem, it just makes the LACP state
machine a little bit more straight forward. For this reason we
decided not to backport it. To be a little bit more specific.
Before this patch:
In a LACP bond, if you unplugged one of the slaves. The bond would
failover as e
Hi Ethan,
I'm looking at 1.4.2 and 1.5.0 releases announced recently and the patches
you've included in the master for this fix are not applied.
Is there any particular reason for that or perhaps the releases were tagged
before you've merged the patch ?
Thanks,
dan
-Original Message-
Thanks, Jesse. Could you explain more details on what features are not
available in the user space datapath?
On Thu, Jun 7, 2012 at 6:05 AM, Jesse Gross wrote:
> On Thu, Jun 7, 2012 at 12:52 PM, Dong Jin wrote:
> > I am running userspace mode of ovs with the following topology
> >
> > VM1(eth0)
On Thu, Jun 7, 2012 at 12:52 PM, Dong Jin wrote:
> I am running userspace mode of ovs with the following topology
>
> VM1(eth0) (eth0)VM2(eth1) --- (eth0)VM3
>
> VM1 and VM3 are end hosts, and VM2 runs the userspace ovs daemon to
> emulate a two-port Ethernet switch.
>
> I sent UDP traffic fr
Hi Justin,
thnx for the explanations.
Here is an excerpt of a scenario, when CPU-load goes up, though within our
network the lost-figures don't normally change:
20120607-114420: lookups: hit:21149788628 missed:12736368714 lost:210746961
flows: 3280 SHORT_FLOWS=2451, TOP=mem: 19m cp
On Jun 6, 2012, at 2:52 AM, Oliver Francke wrote:
> @Justin: Any other recommendations?
Are you also having many short-lived flows? If you're in the range I mentioned
in my response to Kaushal (roughly 120,000 flow setups per second), then the
forthcoming 1.7.0 release may be enough for you.
You shouldn't have to perform any actions on GRE tunnels. You should just be
able to treat it like a physical port. The decapsulation on receive and
encapsulation on send should be invisible to you. They're simply logical ports.
--Justin
On Jun 6, 2012, at 7:17 PM, ravi kerur wrote:
> How
Thanks for the information. We've seen that OVS can handle over 10Gbps. The
problem that you're seeing is related to flow setups. In releases prior to
1.7, the flow setup rate was roughly 40,000 flows per second. The changes in
1.7 increase that number to 120,000.
As we discussed, the bulk
19 matches
Mail list logo
