The VMs are the same, but the traffic is not, one has VLAN tags,
which seems to be provoking a difference in handling.  Check that
your drivers / intermediate switches are handling things correctly.

  -Reid


On Sun, Oct 20, 2013 at 5:57 PM, Li, Chen <chen...@intel.com> wrote:

>  But, I don’t understand, all part of the two VMs are exactly the same
> except the network, because they’re all created by OpenStack.****
>
> The two VM booted on the same system, on the same NIC card, linked to the
> same switch……****
>
> ** **
>
> ** **
>
> *From:* Reid Price [mailto:rpr...@nicira.com]
> *Sent:* Monday, October 21, 2013 7:41 AM
> *To:* Li, Chen
> *Cc:* Jesse Gross; dev@openvswitch.org
>
> *Subject:* Re: [ovs-dev] vlan tag add/remove affect bandwidth a lot****
>
> ** **
>
> Hi chen,****
>
> ** **
>
> What Jesse is pointing out is that your problem seems unrelated to OVS.
>  In particular, the low throughput you observe is probably related to
> something else in the system, be it drivers, VLAN offloading, intermediate
> network, etc.****
>
> ** **
>
> The point is that you already have verified for yourself that the actions
> don't really have an impact, especially compared to the effect when the
> traffic is ultimately tagged =)****
>
> ** **
>
>   -Reid****
>
> ** **
>
> On Sat, Oct 19, 2013 at 4:57 AM, Li, Chen <chen...@intel.com> wrote:****
>
> Sorry, I’m really new to OVS, can you give me more specific move I can do
> next?****
>
> I’m eager to solve this ,but really have no idea where to start……L****
>
>  ****
>
> *From:* Jesse Gross [mailto:je...@nicira.com]
> *Sent:* Saturday, October 19, 2013 12:16 AM
> *To:* Li, Chen
> *Cc:* dev@openvswitch.org
> *Subject:* Re: [ovs-dev] vlan tag add/remove affect bandwidth a lot****
>
>  ****
>
> On Fri, Oct 18, 2013 at 12:08 AM, Li, Chen <chen...@intel.com> wrote:****
>
>  I have changed my test method a little to make the issue more clear:****
>
> ****
>
>  ****
>
> As you can see, my vm has two virtual NIC, tap1 is working under VLAN
> (tag=2002), while tap2 is working under FLAT. ****
>
> And the bandwidth is :****
>
> Test Node => tap1   2.85 Gbits/sec****
>
> Test Node => tap2   9.40 Gbits/sec****
>
>  ****
>
> Flows rules applied:****
>
> ovs-ofctl dump-flows br-int****
>
> NXST_FLOW reply (xid=0x4):****
>
> cookie=0x0, duration=2024.067s, table=0, n_packets=989820,
> n_bytes=23566560107, idle_age=1, priority=3,in_port=16,*vlan_tci=0x0000
> actions=mod_vlan_vid:2*,NORMAL****
>
> cookie=0x0, duration=2024.769s, table=0, n_packets=4873950, n_bytes=
> 7378336782, idle_age=20, priority=3,in_port=17,*dl_vlan=2002
> actions=mod_vlan_vid:1*,NORMAL****
>
> cookie=0x0, duration=2029.604s, table=0, n_packets=52, n_bytes=3768,
> idle_age=72, priority=2,in_port=16 actions=drop****
>
> cookie=0x0, duration=2027.391s, table=0, n_packets=4729, n_bytes=286686,
> idle_age=1, priority=2,in_port=17 actions=drop****
>
> cookie=0x0, duration=2031.156s, table=0, n_packets=3161368,
> n_bytes=208726297, idle_age=20, priority=1 actions=NORMAL****
>
>  ****
>
> ovs-ofctl dump-flows br-eth4****
>
> NXST_FLOW reply (xid=0x4):****
>
> cookie=0x0, duration=2043.307s, table=0, n_packets=925503,
> n_bytes=61092489, idle_age=90, priority=4,in_port=7,*dl_vlan=2
> actions=strip_vlan*,NORMAL****
>
> cookie=0x0, duration=2048.566s, table=0, n_packets=47, n_bytes=3212,
> idle_age=91, priority=2,in_port=7 actions=drop****
>
> cookie=0x0, duration=2049.988s, table=0, n_packets=989979,
> n_bytes=23566580239, idle_age=1, priority=1 actions=NORMAL****
>
>  ****
>
> ovs-ofctl dump-flows br-eth5****
>
> NXST_FLOW reply (xid=0x4):****
>
> cookie=0x0, duration=2047.936s, table=0, n_packets=2235861,
> n_bytes=147635900, idle_age=43, priority=4,in_port=4,*dl_vlan=1
> actions=mod_vlan_vid:2002*,NORMAL****
>
> cookie=0x0, duration=2050.289s, table=0, n_packets=4784, n_bytes=293458,
> idle_age=1, priority=2,in_port=4 actions=drop****
>
> cookie=0x0, duration=2051.95s, table=0, n_packets=305743, n_bytes=
> 7076819244, idle_age=1, priority=1 actions=NORMAL****
>
>  ****
>
> Both path has an adding and stripping of VLAN tags operation, so where
> can I find the difference from code ??****
>
>   ****
>
> The problem is almost certainly not the actual modification of tags but in
> the interactions with the rest of the system. In that case, it's not
> surprising that rewriting the tags doesn't change anything.****
>
>
> _______________________________________________
> dev mailing list
> dev@openvswitch.org
> http://openvswitch.org/mailman/listinfo/dev****
>
> ** **
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev

Reply via email to