Now I removed the tap interfaces, and started from scratch again. However
it doesn't work either. This is what I did:
DBGvpp# set interface state eth1 up
DBGvpp# set interface state eth2 up
DBGvpp# set interface rx-mode eth1 interrupt
DBGvpp# set interface rx-mode eth2 interrupt
DBGvpp# show interface rx-placement
Thread 1 (vpp_wk_0):
node dpdk-input:
eth1 queue 0 (interrupt)
Thread 2 (vpp_wk_1):
eth2 queue 0 (interrupt)
DBGvpp# show hardware-interfaces
DBGvpp#
It is also weird to me that there is no output for 'show
hardware-interfaces'. The NIC is Intel i210 and was loaded with igb driver
before loaded with vfio driver.
Regards,
Xiaodong
On Wed, Jul 20, 2022 at 1:52 PM Pim van Pelt <p...@ipng.nl> wrote:
> Xiaodong,
>
> That tip about LCP is a good one - the tap interfaces you created (or that
> linux-cp created for you) will be in polling mode. virtio/tap does allow
> for adaptive as well. And indeed, even if my physical interfaces (Intel
> X710 10G and Intel i810 100G) are showing adaptive mode, the tap is still
> polling.
>
> For me:
> set int state GigabitEthernet3/0/0 up
> set int state GigabitEthernet3/0/1 up
> set int state HundredGigabitEthernet12/0/0 up
> set int state HundredGigabitEthernet12/0/1 up
>
> set int rx-mode GigabitEthernet3/0/0 adaptive
> set int rx-mode GigabitEthernet3/0/1 adaptive
> set int rx-mode HundredGigabitEthernet12/0/0 adaptive
> set int rx-mode HundredGigabitEthernet12/0/1 adaptive
>
> lcp create GigabitEthernet3/0/0 host-if e0
> set int rx-mode tap1 adaptive
>
> (and ensure all other taps and phys are in adaptive) the load goes from
> 300% to 19% on a machine with 3 workers. So, in your case possibly the tap
> was still being polled?
> groet,
> Pim
>
>
> On Wed, Jul 20, 2022 at 10:43 PM Mohsin Kazmi via lists.fd.io <sykazmi=
> cisco....@lists.fd.io> wrote:
>
>> Hi Xiaodong,
>>
>>
>>
>> What is the output of *show int rx-placement*
>>
>>
>>
>> -br
>>
>> Mohsin
>>
>>
>>
>> *From: *<vpp-dev@lists.fd.io> on behalf of Xiaodong Xu <
>> stid.s...@gmail.com>
>> *Reply-To: *"vpp-dev@lists.fd.io" <vpp-dev@lists.fd.io>
>> *Date: *Wednesday, July 20, 2022 at 8:44 PM
>> *To: *"vpp-dev@lists.fd.io" <vpp-dev@lists.fd.io>
>> *Subject: *[vpp-dev] CPU usage for vpp
>>
>>
>>
>> Hi,
>>
>>
>>
>> When I start vpp, the CPU usage will be close to 100% on the core that
>> vpp is running on. I know this is normal because dpdk is running in poll
>> mode.
>>
>> However when I set the rx-mode to 'interrupt', it doesn't make any
>> difference. The command line I use looks like:
>>
>>
>>
>> # set interface rx-mode [ifname] interrupt
>>
>>
>>
>> Did I miss anything? Thanks for your attention and any suggestions are
>> welcome.
>>
>>
>>
>> Regards,
>>
>> Xiaodong
>>
>>
>>
>>
>>
>>
>
> --
> Pim van Pelt <p...@ipng.nl>
> PBVP1-RIPE - http://www.ipng.nl/
>
>
>
>
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#21700): https://lists.fd.io/g/vpp-dev/message/21700
Mute This Topic: https://lists.fd.io/mt/92512559/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/1480452/21656/631435203/xyzzy
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
