[PATCH 0/2] Latencystat optimization and fix

[Varghese, Vipin]

[Public]

> -----Original Message-----
> From: Stephen Hemminger <stephen at networkplumber.org>
> Sent: Friday, June 13, 2025 6:04 AM
> To: dev at dpdk.org
> Cc: Stephen Hemminger <stephen at networkplumber.org>
> Subject: [PATCH 0/2] Latencystat optimization and fix
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> After investigating latencystat test failures, discovered it poorly designed to handle
> bursts and multiple queues.

Acked-by: Vipin.Varghese at amd.com
Reviewed-by: Vipin.Varghese at amd.com
Tested-by: Thiyagarajan.P at amd.com

We tested the changes using DPDK-TestPmd, following is the summary of the test shared below

1. before vs after patch with 1Q: min: -10ns, max: -5000ns, avg: -10ns, jitter: +5ns
2. before vs after patch with 4Q: min: -10ns, max: -7000ns, avg: +5ns, jitter: +80ns

Since we see a greater variation with multiple queues compared to single queue, we also collect HW Traffic Generator before vs after.
Values we got are ` min: -50000ns, max: -62000ns, avg: -5300ns, jitter: 6ns `.

NIC: Intel E810 2CQ-DA2 (used 1 100Gbps port), firmware: 4.7, DDP: default
Processor: AMD EPYC 8534P 64-Core Processor
Traffic Rate: 100Gbps

Application CMD:
1. DPDK: ` sudo build/app/dpdk-testpmd -l 11,16,17,18,19 -a 42:00.0 --file-prefix tr01 --proc-type=primary --force-max-simd-bitwidth=512 -- --nb-cores=4 --nb-ports=1 --txq=4 --rxq=4 --txd=2048 --rxd=2048 --burst=64 --rss-udp --latencystats=16  -i -a `
2. Proc-Info: ` sudo build/app/dpdk-proc-info --file-prefix=tr01 --proc-type=secondary -a 42:00.0 -- --metrics `

>
> Stephen Hemminger (2):
>   latencystats: fix receive sample MP issues
>   latencystats: optimize locking on transmit
>
>  lib/latencystats/rte_latencystats.c | 64 +++++++++++++++++++----------
>  1 file changed, 42 insertions(+), 22 deletions(-)
>
> --
> 2.47.2