[ovs-discuss] 答复: help - ovs 2.5 flow statistics issues

awang10 at masonlive.gmu.edu awang10 at masonlive.gmu.edu
Fri Aug 12 19:09:13 UTC 2016


Hi, Bhanu,


Thank you for your response! I was trying to repeat the experiment on my machine as well.

Unfortunately, I could still observe the packet loss. One difference I noticed is that I was using linux kernel 3.13.0-92-generic while you were using DPDK.

Does this mean that it only works with DPDK?


Thanks, An

________________________________
发件人: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>
发送时间: 2016年8月11日 22:46:00
收件人: awang10 at masonlive.gmu.edu; discuss at openvswitch.org
主题: RE: help - ovs 2.5 flow statistics issues

Hello An,

I tired reproducing the issue you reported, unfortunately I couldn't reproduce it on my setup. I am using DPDK datapath and OVS Master.
I have increased the number of flows and tried at different rates(made sure no packet drops happen) and found the stats always correct.

Ouput:
cookie=0x0, duration=537.652s, table=0, n_packets=1300000, n_bytes=78000000, idle_age=107, priority=50000,ip,in_port=1,nw_src=2.1.1.38,nw_dst=1.1.25.39 actions=output:2
cookie=0x0, duration=536.761s, table=0, n_packets=3900000, n_bytes=234000000, idle_age=107, in_port=1 actions=output:2

The flow rule set at a much higher priority(50000) to handle packets with above src/dst accounts all the packets correctly and I confirmed from my IXIA settings. All other packets are handled by other flow rule that forwards packets received on port 1 to port2.

Regards,
Bhanu Prakash.

>-----Original Message-----
>From: awang10 at masonlive.gmu.edu [mailto:awang10 at masonlive.gmu.edu]
>Sent: Wednesday, August 10, 2016 2:55 PM
>To: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>;
>discuss at openvswitch.org
>Subject: 答复: help - ovs 2.5 flow statistics issues
>
>Hi, Bhanu,
>
>I could still observe the issue. Please keep me in the loop if you have any
>updates.
>
>Thanks, An
>________________________________________
>发件人: awang10 at masonlive.gmu.edu
>发送时间: 2016年8月4日 1:14:57
>收件人: Bodireddy, Bhanuprakash; discuss at openvswitch.org
>主题: 答复: help - ovs 2.5 flow statistics issues
>
>Hi, Bhanu,
>
>In my case, I don't think it is the rate issue. I sent the traffic at a very low rate
>(~0.7 Mbps) and the statistics are still not correct.
>Also, in my trace for the test, traffic belonging 10.0.0.0/8 -> 81.0.0.0/8 only
>serves a small portion.
>The total statistics combined together are correct, e.g. the results I observed:
>
>NXST_FLOW reply (xid=0x4):
>cookie=0x0, duration=2.183s, table=0, n_packets=12, n_bytes=2083,
>idle_age=2, priority=50000,ip, in_port=1,nw_src=10.0.0.0/8 actions=output:2
>cookie=0x0, duration=10.829s, table=0, n_packets=2328320,
>n_bytes=216980958, idle_age=10, in_port=1 actions=output:2
>
>The sum of 12 and 2328320 is correct. However, there should be 2075 packets
>belonging to 10.0.0.0/8 -> 81.0.0.0/8.
>After looking into the kernel flow table, I found that in the beginning, masked
>key 10.0.0.0/255.0.0.0 -> 81.0.0.0/255.0.0.0 was installed and packets were
>processed fine.
>But after a while, this masked key will be removed by the revalidator
>thread due to the mismatches of the flows in the kernel table.
>So my guess is that either I misconfigured the kernel flow table or there is
>something wrong with the kernel.
>Could you please help further verify this? Appreciate your help!
>
>Thanks,
>An
>
>
>
>________________________________________
>发件人: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>
>发送时间: 2016年8月3日 17:55:44
>收件人: awang10 at masonlive.gmu.edu; discuss at openvswitch.org
>主题: RE: help - ovs 2.5 flow statistics issues
>
>>-----Original Message-----
>>From: awang10 at masonlive.gmu.edu
>[mailto:awang10 at masonlive.gmu.edu]
>>Sent: Monday, August 1, 2016 3:49 PM
>>To: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>;
>>discuss at openvswitch.org
>>Subject: 答复: help - ovs 2.5 flow statistics issues
>>
>>Thank you, Bhanu, for your verification. I really appreciate it!
>>Please keep me updated if you figure out the root cause of this problem.
>
>Can you reduce the packet rate and reverify the stats. I don’t see a problem
>with stats when adjusted my packet rate.
>
>Regards,
>Bhanu Prakash.
>
>>
>>Thanks,  An
>>________________________________________
>>发件人: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>
>>发送时间: 2016年8月1日 22:43:37
>>收件人: awang10 at masonlive.gmu.edu; discuss at openvswitch.org
>>主题: RE: help - ovs 2.5 flow statistics issues
>>
>>>-----Original Message-----
>>>From: awang10 at masonlive.gmu.edu
>>[mailto:awang10 at masonlive.gmu.edu]
>>>Sent: Monday, August 1, 2016 12:52 PM
>>>To: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>;
>>>discuss at openvswitch.org
>>>Subject: 答复: help - ovs 2.5 flow statistics issues
>>>
>>>Hi, Bhanu,
>>>
>>>Thank you for your response! I tried it with the master branch and it works!
>I
>>>guess it is a problem with the 2.5.0 version.
>>>But I noticed that the statistics are actually wrong in my case even though I
>>>could see the counts.
>>
>>You are right, the stats doesn't seem accurate in my case too.
>>I tested it by sending 100k packets for 5 runs and below is  'n_packets' count
>in
>>dump-flows output.
>>
>> Reported | Actual
>>97206         | 100,000
>>97833         | 100,000
>>97954         | 100,000
>>98334         | 100,000
>>98017         | 100,000
>>
>>Regards,
>>Bhanu Prakash.
>>
>>>Could you please help verify if you have obtained accurate results for your
>>>case?
>>>
>>>Thanks,
>>>An
>>>________________________________________
>>>发件人: Bodireddy, Bhanuprakash <bhanuprakash.bodireddy at intel.com>
>>>发送时间: 2016年8月1日 15:59:13
>>>收件人: awang10 at masonlive.gmu.edu; discuss at openvswitch.org
>>>主题: RE: help - ovs 2.5 flow statistics issues
>>>
>>>>-----Original Message-----
>>>>From: discuss [mailto:discuss-bounces at openvswitch.org] On Behalf Of
>>>>awang10 at masonlive.gmu.edu
>>>>Sent: Wednesday, July 27, 2016 3:14 PM
>>>>To: discuss at openvswitch.org
>>>>Subject: [ovs-discuss] help - ovs 2.5 flow statistics issues
>>>>
>>>>Hi, All,
>>>>
>>>>I am running openvswitch 2.5.0 on Ubuntu 14.04 LTS with kernel 3.13.0-92-
>>>>generic.
>>>>I installed the two following rules to the switch:
>>>>
>>>>NXST_FLOW reply (xid=0x4):
>>>> cookie=0x0, duration=2.183s, table=0, n_packets=0, n_bytes=0,
>>idle_age=2,
>>>>priority=50000,ip,in_port=1,nw_src=10.0.0.0/8,nw_dst=81.0.0.0/8
>>>>actions=output:2
>>>> cookie=0x0, duration=10.829s, table=0, n_packets=0, n_bytes=0,
>>>>idle_age=10, in_port=1 actions=output:2
>>>>
>>>>Another host machine connecting to the switch generates traffic
>(including
>>>>10.0.0.0/8 -> 81.0.0.0/8) that goes through the switch.
>>>>However, the statistics were not be updated correctly with n_packets for
>>>the
>>>>higher priority rule remaining 0 while all the traffic statistics being credited
>>to
>>>>the lower priority rule.
>>>>But if I modify the rules to be:
>>>>
>>>>NXST_FLOW reply (xid=0x4):
>>>> cookie=0x0, duration=2.183s, table=0, n_packets=0, n_bytes=0,
>>idle_age=2,
>>>>priority=50000,ip,in_port=1,nw_src=10.0.0.0/8 actions=output:2
>>>> cookie=0x0, duration=10.829s, table=0, n_packets=0, n_bytes=0,
>>>>idle_age=10, in_port=1 actions=output:2
>>>>
>>>>Then it works fine. So it seems that statistics with both nw_src and nw_dst
>>>>matching do not work properly?
>>>
>>>I don't see an issue matching both nw_src & nw_dst and can see the
>>>'n_packets', 'n_bytes' entry incremented for all matching packets. I quickly
>>>verified this on master branch.
>>>
>>>NXST_FLOW reply (xid=0x4):
>>> cookie=0x0, duration=430.063s, table=0, n_packets=2109130027,
>>>n_bytes=126547801620, idle_age=9,
>>>priority=50000,ip,in_port=1,nw_src=2.0.0.0/8,nw_dst=3.0.0.0/8
>>>actions=output:2
>>> cookie=0x0, duration=428.842s, table=0, n_packets=0, n_bytes=0,
>>>idle_age=428, in_port=1 actions=output:2
>>>
>>>Regards,
>>>Bhanu Prakash.
>>>
>>> Or did I misconfigure something?
>>>>I'd really appreciate your help!
>>>>
>>>>Thanks,
>>>>An
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160812/8504bce8/attachment-0002.html>


More information about the discuss mailing list