We have a platform as shown below:
Now we run RFC2544 throughput test for Linux native ip forwarding, and get the result as show below:
Focus on 1518B, the throughput is totally 100%.
However, we run RFC2544 test with DPDK (testpmd), and here is the result.
(My command is: # ./testpmd -c 0x3 -n 1 -- -i --nb-cores=1 --nb-ports=2 --total-num-mbufs=2048)
The throughput of 1518B is just only 99.297%,
which means it is failed when running 100% offered loading because there is a few frame loss (~0.5%).
Could anyone help us to explain the phenomenon?
Could anyone help us to improve it to 100%?
BTW, we have been search this issue, and I found this post
I just attempted to modify these parameters like rxd, txd, txwt, burst, etc,.. but it doesn't seem work to me.
OS: CentOS 7.1 x64
DPDK ver: 16.04
Thank you for contacting Intel Embedded Community.
In order to better understand this situation, could you please try to reproduce this condition with any of the Operating System (OS) listed on page 3 of the http://www.intel.com/content/dam/www/public/us/en/documents/platform-briefs/atom-processor-e3800-platform-brief.pdf Bay Trail-I Product Brief and let us know the outcomes in a detailed way?
Also, could you please confirm if you have addressed your questions to the emails listed at the http://dpdk.org/ml DPDK Mailing List?
By the way, please review as a reference that your implementation fulfills with the guidelines stated at the http://dpdk.org/doc/guides-1.8/testpmd_app_ug/index.html Testpmd Application User Guide.
Thanks in advance for your help to solve this inconvenience.