[PATCH] app/testpmd:add vxlan txonly

[Ferruh Yigit]

On 2/18/2023 11:06 AM, wushaohua wrote:
> 
> 在 2023/2/18 4:43, Ferruh Yigit 写道:
>> On 1/3/2023 2:30 AM, wushaohua at chinatelecom.cn wrote:
>>> From: Shaohua Wu <wushaohua at chinatelecom.cn>
>>>
>>> Add the vxlan packet sending module to actively send
>>> vxlan packets using a common network adapter.
>>> The default vni is 1000.
>>> example:
>>>  ./dpdk-testpmd -l 32-47 -n 16 --file-prefix pg0 -- -i
>>> 	--rxq=16 --txq=16 --rxd=1024 --txd=1024
>>> 	--txpkts=64 --burst=64  --mbuf-size=4096
>>> 	--nb-cores=15  --underlay_tx_only
>>> 	--underlay-eth-peer=0,f0:00:00:00:00:66
>>> 	--eth-peer=0,08:c0:eb:3e:87:af
>>> 	--utx-ip=11.0.0.1,11.0.0.2
>>> 	--tx-ip=30.0.0.1,30.0.0.2
>>> 	--forward-mode=tuntxonly
>>> 	--txonly-multi-flow
>>>
>>> Signed-off-by: Shaohua Wu <wushaohua at chinatelecom.cn>
>> Hi Shaohua,
>>
>> This is a good feature but I am not sure if this is in the scope of
>> testpmd application.
>>
>> Testpmd has basic packet generation capabilities, and it is useful for
>> quick/dirty testing, but for more advanced packet generation perhaps
>> other tools like pktgen [1] can be used, what do you think?
>>
>> What is the justification to have this feature in testpmd, and what is
>> your usecase?
>>
>>
>> [1]
>> https://git.dpdk.org/apps/pktgen-dpdk/
> 
> Hi Ferruh，
> 
> Thank you for your review.From what I know now pktgen doesn't seem to support the construction of vxlan packets.
> 

I was asking if it makes sense to add this support to pktgen, instead of
testpmd.

> In a real test scenario, testpmd only sends bare packets and cannot cover test scenarios, such as ovs decap performance.
> Currently, testpmd of this patch is used, and the packet sending performance is 22+Mpps with a 25G network adapter.
> 
> This is necessary for testing DPU devices or virtualization scenarios.
>