Performance degradation during dpdk source code testing

     There is no principle in this article , Make a note of a small problem I encountered , For your reference .
    DPDK There are many ways to test the performance of , What I use here is compiled from the source code testpmd Program .
     I have the following problems ：

1. The hardware configurations of the two machines involved in the test are identical ,BIOS Same version , The kernel version is basically the same , The various configurations of the test program are identical , But the performance of one machine is much worse than that of the other , The gap is 50% above ;
2. testpmd The package received seen in pps and bps The values do not match , such as , Package sizes are 64 byte ,testpmd Given pps yes 5582583, So the corresponding bps Should be 2858282496, and testpmd Given bps yes 9318213000, Two phase comparison , Great difference ;
3. The sender uses pktgen Set the packet size to 400 After byte , All packets at the receiving end are lost ;

     Above 3 The first question is very confusing . Although the hardware is normal 、 Correct use , But the test results are problematic . I haven't analyzed the reason , Finally, all the files generated during the previous build are deleted directly , Then rebuild , The problem was solved . From this point of view , It may be that different versions have been built in the same directory many times before DPDK And cause problems . The specific reasons are difficult to analyze , If you are using DPDK When performing performance tests on the source code , Encountered a similar strange problem , Consider rebuilding DPDK.