Trex with mlx5 driver - Multiple streams with different VLAN priority causes high CPU utilization

[Rubens Figueiredo <rubens.figueiredo@bisdn.de>]

[-- Attachment #1: Type: text/plain, Size: 3043 bytes --]

Hello community,

I am facing a strange issue in the Trex stateless code, version v3.02 
and v3.04. I am using the Mellanox Cx-5, and have created two VFs on top 
of the PF 0. The mlx5_core version I am using is the 5.7-1.0.2, and the 
ofed version is MLNX_OFED_LINUX-5.7-1.0.2.0 (OFED-5.7-1.0.2).

I have created the following issue in the trex-core repository 
[here](https://github.com/cisco-system-traffic-generator/trex-core/issues/1124), 
and was recommended to post the issue in here. In the github issue you 
see screenshots of the issue I am facing.

I am trying to create two parallel streams with different VLAN 
priorities, but the load generated is not what I expect it to be, and 
CPU util. seems incredibly high (~99%).

I have reproduced this issue with the --software and non software version.

The script I used is below.

|import stl_path from trex.stl.api import * import time import pprint 
from ipaddress import ip_address, ip_network import argparse import 
configparser import os import json def get_packet(tos, mac_dst, ip_src, 
size): # pkt = Ether(src="02:00:00:00:00:01",dst="00:00:00:01:00:01") / 
IP(src="10.0.0.2", tos=tos) / UDP(sport=4444, dport=4444) pkt = ( 
Ether(src="00:01:00:00:00:02", dst=mac_dst) # 
Ether(dst="11:11:11:11:11:11") # / Dot1AD(vlan=0) / Dot1Q(vlan=0, 
prio=tos) / IP(src=ip_src) / UDP(sport=4444, dport=4444) ) pad = max(0, 
size - len(pkt)) * "x" return pkt / pad def main(): """ """ tx_port = 0 
rx_port = 1 c = STLClient() # connect to server c.connect() # prepare 
our ports c.reset(ports=[tx_port, rx_port]) streams = [] s = STLStream( 
packet=STLPktBuilder( pkt=get_packet(4,"00:11:22:33:44:55", 
"10.1.0.2",512), # vm = vm, ), isg=0 * 1000000, 
mode=STLTXCont(pps=1.2*10**6), # flow_stats = STLFlowLatencyStats(pg_id 
= 0) flow_stats = STLFlowStats(pg_id=0), ) streams.append(s) s2 = 
STLStream( packet=STLPktBuilder( pkt=get_packet(2,"00:11:22:33:44:55", 
"10.1.0.2",512), # vm = vm, ), isg=0 * 1000000, 
mode=STLTXCont(pps=1.2*10**6), # flow_stats = STLFlowLatencyStats(pg_id 
= 0) flow_stats = STLFlowStats(pg_id=1), ) streams.append(s2) 
c.add_streams(streams, ports=[tx_port]) c.clear_stats() 
c.start(ports=[tx_port], duration=60, mult="25gbpsl1") 
c.wait_on_traffic(ports=[tx_port, rx_port]) stats = c.get_stats() 
print(stats) if __name__ == "__main__": main() And the configuration is |

|- port_limit: 2 version: 2 port_bandwidth_gb: 100 interfaces: 
["3b:00.2", "3b:00.3"] port_info: - dest_mac: 00:00:00:00:00:01 src_mac: 
00:01:00:00:00:01 - dest_mac: 00:00:00:00:00:02 src_mac: 
00:01:00:00:00:02 c: 14 platform: master_thread_id: 8 latency_thread_id: 
27 dual_if: - socket: 0 threads: 
[9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26]|


-- 
BISDN GmbH
Körnerstraße 7-10
10785 Berlin
Germany


Phone: 
+49-30-6108-1-6100


Managing Directors: 
Dr.-Ing. Hagen Woesner, Andreas 
Köpsel


Commercial register: 
Amtsgericht Berlin-Charlottenburg HRB 141569 
B
VAT ID No: DE283257294


[-- Attachment #2: Type: text/html, Size: 5679 bytes --]