[ovs-dev] [PATCH v2] netdev: Custom statistics.

[Kevin Traynor]

On 12/19/2017 03:00 PM, Kevin Traynor wrote:
> On 12/19/2017 02:35 PM, Weglicki, MichalX wrote:
>>> -----Original Message-----
>>> From: Kevin Traynor [mailto:ktraynor at redhat.com]
>>> Sent: Tuesday, December 19, 2017 3:07 PM
>>> To: Weglicki, MichalX <michalx.weglicki at intel.com>; dev at openvswitch.org
>>> Subject: Re: [ovs-dev] [PATCH v2] netdev: Custom statistics.
>>>
>>> On 12/05/2017 02:55 PM, Michal Weglicki wrote:
>>>> - New get_custom_stats interface function is added to netdev. It
>>>>   allows particular netdev implementation to expose custom
>>>>   counters in dictionary format (counter name/counter value).
>>>> - New statistics are retrieved using experimenter code and
>>>>   are printed as a result to ofctl dump-ports.
>>>> - New counters are available for OpenFlow 1.4+.
>>>> - New statistics are printed to output via ofctl only if those
>>>>   are present in reply message.
>>>> - New statistics definition is added to include/openflow/intel-ext.h.
>>>> - Custom statistics are implemented only for dpdk-physical
>>>>   port type.
>>>> - DPDK-physical implementation uses xstats to collect statistics.
>>>>   Only dropped and error counters are exposed.
>>>>
>>> Hi Michal - why only dropped and error counters? why not just expose
>>> them all. For example, IIUC this would report management dropped packets
>>> but there would not be a stat for management rx/tx successful packets.
>>>
>>> Kevin.
>> Hi Kevin - those counters were of biggest value to us at the point of making this 
>> patch, sending all counters (where for IXGBE is about 150) will produce 
>> some movement on the network. I think that biggest advantage of this 
>> particular patch is that it introduces a mechanism to expose 
>> any counters, counters list can be extended in the future 
>> if necessary. However I'm not sure if sending all counters 
>> is good idea, as there could be thousands of it in the future - in this 
>> solution, we have some kind of control over the data size. 
>>
> Ok thanks, that makes sense. I would like to suggest that *_management_*
> be added as part of this as I think it's only 2 additional stats and
> I've seen at least one user saying they needed this information.
> 

Hi Michal, this does not look to be in v3, do you think it should be
added as a separate patch? or they should not be reported?

thanks,
Kevin.