Hi, can CX6/CX7 can rate limit hairpin queue? Because too much hairpin flow will influence normal TX rate.
Will this hairpin rate limit occurs packet loss?
Hi ,
I suggest to open support ticket for this question by sending email to enteprisesupport@nvidia.com
In order to understand the case scenario .
In general , Mellanox/NVIDIA ConnectX-6 (CX6) and ConnectX-7 (CX7) network adapters support advanced queueing and Quality of Service (QoS) features, including rate limiting at the queue level. However, directly rate limiting hairpin queues (used for forwarding packets between ports or between queues on the same card without involving the host CPU) is not straightforward and depends on how traffic is classified and prioritized.
Recent discussions in the DPDK community show that while you can use QoS mechanisms to control bandwidth allocation between different traffic classes (e.g., hairpin vs. normal TX), you cannot directly use flow rules to assign priority to hairpin traffic alone. Instead, you must mark packets with VLAN PCP (Priority Code Point) or IPv4 DSCP (Differentiated Services Code Point) values, and then configure the NIC to allocate bandwidth or buffer resources according to these priorities1. This approach allows you to effectively “rate limit” hairpin traffic by controlling how much bandwidth or buffer it can consume relative to other traffic.
Thanks
Samer