Queue_full and virtio vhu interface

[Haresh Khandelwal]

Hi,

I am running Trex on VM (RHEL 9.4) with 1 core and 1 interface via VHU port attached to OVS user bridge. The physical interface attached to the OVS-DPDK bridge is a MLX card. 

Now with 0.75Mpps traffic sent, I am noticing a significant queue full (snapshot attached). I do understand there is back pressure, however I see this behaviour with all egress traffic rate. 

For example, 

when I send 1.5 mpps I see around 200 packets per sec in queue_full while I do see tx pps is constant at 1.5 mpps.

When I send 1 mpps, I see same amount of packets filling in queue and tx rate is constant at 1 mpps.

I did try increasing the Tx and Rx queue descriptor size of the VHU port to 4096 on ovs. But I didn't notice any difference.

Is there any known behaviour with trex using virto (vhu interface) doing something like this? Any way to get away from this?

Regards,

Haresh

[Haresh Khandelwal]

Hi folks, 

Is there anyone who has observed something like this before or any comment on what we are trying!

Thanks

-Haresh

[Цветков Андрей]

Hi 
Set addishinal pameters for virt nic -  queues="16" rx_queue_size="1024" tx_queue_size="256". 
add  to vm more cores . add cores to trex

are trex run in hardware mode ? config["low_end"] = False

<interface type="network">
  <mac address="52:54:00:cd:b2:d8"/>
  <source network="network-1"/>
  <model type="virtio"/>
  <driver name="vhost" txmode="iothread" ioeventfd="on" event_idx="off" queues="16" rx_queue_size="1024" tx_queue_size="256">
    <host csum="on" gso="on" tso4="on" tso6="on" ecn="off" ufo="off" mrg_rxbuf="off"/>
    <guest csum="on" tso4="on" tso6="on" ecn="off" ufo="off"/>
  </driver>
  <address type="pci" domain="0x0000" bus="0x00" slot="0x0a" function="0x0"/>
</interface>

суббота, 8 февраля 2025 г. в 13:37:12 UTC+3, Haresh Khandelwal: