Hi Try adding -m ./t-rex-64 -i -c 14 --cfg /etc/trex_cfg.yaml -m 100000
/Jimmy From: [email protected] <[email protected]> on behalf of Vladimir Smirnov via lists.fd.io <[email protected]> Date: Wednesday, 14 October 2026 at 17:21 To: [email protected] <[email protected]> Subject: Re: [vpp-dev] Regarding 400gig throughput testing What you've described sounds like one of the example scapy scripts they ship with - it is basically generating every single packet on demand and you are bound by how fast scapy can do the job (hence the CPU utilization). Try to make the script pre-generate more packets in a single go. I used https://github.com/Civil/simple_bench/blob/main/profiles/simple.py this script in the tests and mostly https://github.com/Civil/simple_bench/blob/main/profiles/simple.py#L58 makes it much faster. As I've said, I don't think it is a t-rex limitation, just the default settings that are suboptimal. Of course there might be something that is related to a different driver (I was using mlx5 - so mellanox and nvidia connectx and bluefield cards in my tests, I don't have any broadcom NICs), but that sounds extremely unlikely to me. On Wed, 14 Oct 2026 at 16:50, Dinesh via lists.fd.io<http://lists.fd.io/> <[email protected]<mailto:[email protected]>> wrote: Hello vladimir, It's 24 core machine without hyperthreading. LnkCap: Port #0, Speed 32GT/s, Width x16, ASPM not supported LnkSta: Speed 32GT/s (ok), Width x16 (ok) model name : Intel(R) Xeon(R) Gold 5412U stepping : 7 microcode : 0x2b000161 cpu MHz : 2100.000 memory - DDR5 4800MT Broadcom 400g NIC. Can we modify trex source to get 400 gig with single port ? Also i am giving 14 cores for trex, ./t-rex-64 -i -c 14 --cfg /etc/trex_cfg.yaml Still when i am passing ipv6 udp traffic, cpu utilization is 100% and i cannot push beyond 120gig. On 14/10/26 18:53, Vladimir Smirnov via lists.fd.io<http://lists.fd.io/> wrote: What hardware have you used for T-Rex? About 1.5 years ago I managed to get about ~400 Mpps / ~800 Gbps on a single 16 core machine (https://nog.fi/event/26/contributions/191/attachments/120/262/Generating%201%20Tbps.pdf - slide 39, it was 128b packets), same machine could do about 600 Mpps with 64 byte packets and you can tune the packet size to get around 600 Mpps. Frankly speaking, my NICs on those boxes were capable of 200G per port only. On Wed, 14 Oct 2026 at 13:12, Dinesh via lists.fd.io<http://lists.fd.io/> <[email protected]<mailto:[email protected]>> wrote: Hello everyone, What is the best tool for linux to test 400 gbps throughput in vpp ? My setup is like below. Lan -> Vpp -> Router (Pumping traffic from lan host to router) I want to test both ipv4 and ipv6. I tried with dpdk-testpmd and it works good with ipv4, but there is no support for ipv6 i guess. I also tried with T-rex , but somehow t-rex source code is hard-capped at 200gbps. Thanks, Dinesh -- Best regards, Vladimir Smirnov. -- Best regards, Vladimir Smirnov.
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#27275): https://lists.fd.io/g/vpp-dev/message/27275 Mute This Topic: https://lists.fd.io/mt/121748576/21656 Group Owner: [email protected] Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
