Hi Han,
I am using Openstack with vXLAN & see the performance 2.x Gbps throughput
over the 10Gbps line. I see in the below link that you were facing the same
issue & after tuning some parameters, you could go upto 6Gbps. It would be
great if you can share those tuning parameters? I have configured
Hi Han,
Thanks for your input - much appreciated.
I will try and reproduce their results.
Regards
Kristoffer
On 31/07/2014, at 11.23.07, Han Zhou wrote:
> Hi Kristoffer,
>
> Sorry for late response.
>
> On Tue, Jul 29, 2014 at 4:30 PM, Kristoffer Egefelt
> wrote:
>> I actually asked them:
Hi Kristoffer,
Sorry for late response.
On Tue, Jul 29, 2014 at 4:30 PM, Kristoffer Egefelt wrote:
> I actually asked them:
> http://lists.opencontrail.org/pipermail/users_lists.opencontrail.org/2014-July/000338.html
>
> As I understood it, they are able to offload from VM to VM, without
> segm
I actually asked them:
http://lists.opencontrail.org/pipermail/users_lists.opencontrail.org/2014-July/000338.html
As I understood it, they are able to offload from VM to VM, without
segmentation - maybe I’m wroing, and what actually is happening is that their
vrouter module, on the receiving sid
Hi,
>
> OK - I see, thanks for clearing that up.
> Do you know if VXLAN has the same limitation ?
> The Contrail vrouter seems to be able to work around it..
> http://opencontrail.org/evaluating-opencontrail-virtual-router-performance/
>
Yes, VXLAN is the same. I just read the OpenContrail link yo
Hi Han,
>> GRE: 9000 bytes
>> NIC: 9000 bytes
>>
> Where did you get NIC: 9000 bytes? What's the MTU of your physical interface?
tcpdump on the physical interface show GRE packets are segmented to ~9000 if
the VMs vif has MTU 9000. The MTU on the physical NIC is 9192.
> Otherwise it is normal
Hi Kristoffer,
On Mon, Jul 28, 2014 at 5:29 PM, Kristoffer Egefelt wrote:
> GRE: 9000 bytes
> NIC: 9000 bytes
>
Where did you get NIC: 9000 bytes? What's the MTU of your physical interface?
Otherwise it is normal. TSO of virtual driver of VM offload 64k
buffers to host, and host (without GRE off
Do you mean that GRE offloading is needed for OVS to forward GSO/GRO packets ?
The funny thing is, that it seems to be the VMs MTU that determines what the
GRE MTU will be.
So at VM mtu 1500, packet sizes are:
VM: 64K
VIF: 64K
Bridge: 64K
GRE: 1500 bytes
NIC: 1500 bytes
At VM mtu 9000, packet si
If you are talking about why you are getting ~2 Gbps and not 10 Gbps
using GRE tunnel, then I think it is expected. I haven't gotten more
than 4 Gbps using GRE tunnel (with a lot of network optimizations like
setting affinity etc). If you expect to get 10 Gbps, you should use a
tunnel that has some
I just confirmed that without the GRE tunnel, it works.
So it seems the problem is in the GRE handling in or after openvswitch.
I’m not sure if this is actually supposed to work with GRE packets - do the NIC
drivers need to support GRE offload for this to work?
Or could it be a kernel issue - as
Hi Flavio,
> On Thu, Jul 24, 2014 at 01:47:32PM +0200, Kristoffer Egefelt wrote:
> [...]
>> - Packets egressing openvswitch over a gre tunnel is segmented to
>> 1500 bytes. If I configure mtu 9000 in the VM, the packets are
>> segmented to 9000 bytes on the gre tunnel - in any case tcpdump in
>> t
On Thu, Jul 24, 2014 at 01:47:32PM +0200, Kristoffer Egefelt wrote:
[...]
> - Packets egressing openvswitch over a gre tunnel is segmented to
> 1500 bytes. If I configure mtu 9000 in the VM, the packets are
> segmented to 9000 bytes on the gre tunnel - in any case tcpdump in
> the sending VM shows
Hi,
I’m looking into how this is possible
http://opencontrail.org/evaluating-opencontrail-virtual-router-performance/
achieving 10G line rate between two VMs on different hypervisors over a tunnel
using MTU 1500 in the VMs.
I see (at least) two possible issues with my setup:
- Packets egressi
13 matches
Mail list logo
