That’s the secondary issue which we will focus later on.
So let me rephrase the problem statements -
1. This does not involve zc! Problem 1- When we are using our software integrated with pfring 7.5.0, we do not see counters incremented in cat /proc/interrupts/ for the interface from which we are reading the packets.
We are using standard i40e. driver on rhel 7.6 and have reduced the number of queues of this card interfaces to 4 and set the smp affinity explicitly.
Second problem - we can see ksoftirq for the assigned cores taking 100% of CPU due to which we are observing drop at the interface as well.
When we run tcpdump on the same interface, we can see the interrupts counts incrementing using above same command.
2- above observation is same when we deploy zc compiled i40e driver ;
Will there be no interrupts with i40e driver when used with pf_ring ??
Regards,
Chandrika
Sent from my iPhone
> On Sep 4, 2019, at 1:04 PM, Alfredo Cardigliano <cardigliano@ntop.org> wrote:
>
> Hi Chandrika
> are you observing better performance with ixgbe? Please note that ixgbe has a bigger buffer in the card,
> this could explain better performance in case of high traffic rate (what is the pps rate in this case?)
>
> Alfredo
>
>> On 28 Aug 2019, at 12:56, Chandrika Gautam <chandrika.iitd.rock@gmail.com> wrote:
>>
>> Hi Alfredo,
>>
>> Interrupts are not seen either with standard driver or with the zc compiled i40e driver as well.
>>
>> What we do in our application is -
>> We load zc compiled i40e driver and put a zc license on that interface but we do not open device using zc: prefix.
>> Using this we have observed better performAnce in our application when used with ixgbe driver.
>>
>> Regards,
>> Chandrika
>>
>> Sent from my iPhone
>>
>>> On Aug 28, 2019, at 12:58 PM, Alfredo Cardigliano <cardigliano@ntop.org> wrote:
>>>
>>> Hi Chandrika
>>> are you using the i40e-zc driver or the standard driver?
>>> Please note that ZC enables interrupts only when required for performance reason,
>>> for example some libpcap-based applications require interrupts as they use poll/select,
>>> instead many pf_ring-based application do not use interrupts at all (they do active polling).
>>>
>>> Alfredo
>>>
>>>> On 28 Aug 2019, at 08:59, Chandrika Gautam <chandrika.iitd.rock@gmail.com> wrote:
>>>>
>>>>
>>>> Hi Team,
>>>>
>>>>
>>>>
>>>> We have compiled our software with PF_RING 7.5.0 which reads from an interface on i40e driver.
>>>>
>>>> Whenever we start our software, it process the traffic received on the interface but no interrupts are seen.
>>>>
>>>> Whenever we start tcpdump on the same interface, then interrupts can be seen.
>>>>
>>>> Can you please help to validate this behavior. Is it expected behavior ? We need to validate that the interrupts are coming on the cores to which we have set the affnity.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Regards,
>>>>
>>>> Chandrika
>>>>
>>>> _______________________________________________
>>>> Ntop-misc mailing list
>>>> Ntop-misc@listgateway.unipi.it
>>>> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>>>
>>> _______________________________________________
>>> Ntop-misc mailing list
>>> Ntop-misc@listgateway.unipi.it
>>> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>> _______________________________________________
>> Ntop-misc mailing list
>> Ntop-misc@listgateway.unipi.it
>> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>
> _______________________________________________
> Ntop-misc mailing list
> Ntop-misc@listgateway.unipi.it
> http://listgateway.unipi.it/mailman/listinfo/ntop-misc