[dpdk-users] Packet Drop in same NUMA

[dpdk-users] Packet Drop in same NUMA

[Dheeraj Dang]

Hi all,

I m facing one issue in which packets are continuously dropped at vhost
level in same numa configuration.

*PMD Thread Core Affinity*
pmd thread numa_id 1 core_id 17:
isolated : false
port: dpdk0 queue-id: 0
port: vhu43378b38-cb queue-id: 0
port: vhu8d094814-25 queue-id: 0


*Case I: When receiver application on VM is using numa_id 0 and pmd_thread
is using numa_id 1*

Packet is going from vhost to VM via QPI

Intel(r) QPI traffic estimation in bytes (data and non-data traffic
outgoing from CPU/socket through QPI links):

               QPI0   QPI1
---------------------------------------------------------------------------------------------------------------
 SKT    0       3%     4%
 SKT    1       3%     3%

In this case, no packet is dropped at vhost level.

*Case II: When my receiver application is using numa_id 1 and pmd_thread is
using numa_id 1*

QPI monitoring
Intel(r) QPI traffic estimation in bytes (data and non-data traffic
outgoing from CPU/socket through QPI links):

                 QPI0   QPI1
---------------------------------------------------------------------------------------------------------------
 SKT    0          0%     0%
 SKT    1         0%     0%

In this case, I m facing continuous packet drop at vhost level.

Following things that I tried, but none of them worked
1. RX descriptors values (512, 1024, 2048, 4096)
2. MBUF_CACHE_SIZE values (128, 512)

Please suggest where I m doing wrong? Maybe I've missed something and
compute node is missing some configuration that is causing this issue.

Regards
Dheeraj Dang