You are not logged in.

#1 2023-05-25 05:30:10

girishN23
Member
Registered: 2023-05-25
Posts: 1

Scale: register a network device(register_netdev) takes long time and

we have a network application in which we make netlink call and in the kernel-module we call register_netdev to create a network device we need. This works without any issues for small numbers.

When we create 8k network devices all at once, I could see it takes 2 minutes approx to complete, after which CPU usage continues to be high, mainly with many daemons named systemd-udevd and ifquery

root@NetwAA:~# pgrep -l systemd
1925 systemd-journal
1946 systemd-udevd
2130 systemd-logind
13381 systemd-udevd
13385 systemd-udevd
13393 systemd-udevd
13399 systemd-udevd
13405 systemd-udevd
13411 systemd-udevd
13415 systemd-udevd
13423 systemd-udevd
13428 systemd-udevd
13432 systemd-udevd
13441 systemd-udevd
13446 systemd-udevd
13452 systemd-udevd
13459 systemd-udevd
13462 systemd-udevd
13468 systemd-udevd
13475 systemd-udevd
13483 systemd-udevd
...

root@NetwAA:~# pgrep -l ifquery
13388 ifquery
13392 ifquery
13397 ifquery
13404 ifquery
13409 ifquery
13417 ifquery
...

top - 00:55:05 up  1:14,  2 users,  load average: 70.09, 30.47, 12.17
Tasks: 462 total,  12 running, 450 sleeping,   0 stopped,   0 zombie
%Cpu(s): 19.0 us, 74.1 sy,  0.0 ni,  6.8 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem :  15937.0 total,  13024.0 free,   2283.3 used,    629.7 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.  13308.5 avail Mem

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
1946 root      20   0   26412  11004   3896 R  99.3   0.1   2:22.81 systemd-u+
13551 root      20   0    7660   6820   1556 D  36.5   0.0   0:12.04 ifquery
13564 root      20   0    5812   4816   1404 R  15.6   0.0   0:10.69 ifquery
13656 root      20   0    2908   1968   1460 D  12.0   0.0   0:10.28 ifquery
13590 root      20   0    4096   3396   1564 D  11.6   0.0   0:10.29 ifquery
13711 root      20   0    3568   2856   1556 D  11.3   0.0   0:10.49 ifquery
13572 root      20   0    3964   3104   1540 D  11.0   0.0   0:12.23 ifquery
13784 root      20   0    4096   3236   1404 R  11.0   0.0   0:10.94 ifquery
13471 root      20   0    2380   1812   1564 D  10.6   0.0   0:10.91 ifquery
13640 root      20   0    8848   8020   1440 D  10.6   0.0   0:10.65 ifquery
13701 root      20   0    8056   7300   1512 D  10.0   0.0   0:11.42 ifquery
13630 root      20   0    3304   2592   1556 D   9.6   0.0   0:11.44 ifquery
13704 root      20   0    4492   3640   1548 D   9.6   0.0   0:10.96 ifquery
13392 root      20   0    9112   8304   1460 D   9.0   0.1   0:11.99 ifquery
13521 root      20   0    7660   6704   1440 D   9.0   0.0   0:12.84 ifquery
13770 root      20   0    8716   7768   1452 D   9.0   0.0   0:12.06 ifquery

High CPU usage by systemd-udevd and ifquery

after a long time: //about 10-15 mins

root@NetwAA:~# udevadm monitor
UDEV  [538.005910] add      /devices/virtual/net/xe9.924/queues/tx-0 (queues)
UDEV  [538.011380] add      /devices/virtual/net/xe9.934/queues/rx-0 (queues)
UDEV  [538.248471] add      /devices/virtual/net/xe9.933 (net)
UDEV  [538.266858] add      /devices/virtual/net/xe9.976 (net)
UDEV  [538.358951] add      /devices/virtual/net/xe9.933/queues/rx-0 (queues)
UDEV  [538.359507] add      /devices/virtual/net/xe9.933/queues/tx-0 (queues)
UDEV  [538.359594] add      /devices/virtual/net/xe9.936/queues/rx-0 (queues)
UDEV  [538.360094] add      /devices/virtual/net/xe9.934/queues/tx-0 (queues)
UDEV  [538.360121] add      /devices/virtual/net/xe9.936/queues/tx-0 (queues)
UDEV  [538.609054] add      /devices/virtual/net/xe9.954 (net)
UDEV  [538.624835] add      /devices/virtual/net/xe9.968 (net)
UDEV  [538.797069] add      /devices/virtual/net/xe9.932 (net)
UDEV  [538.804163] add      /devices/virtual/net/xe9.954/queues/rx-0 (queues)
UDEV  [538.813290] add      /devices/virtual/net/xe9.954/queues/tx-0 (queues)
UDEV  [538.813778] add      /devices/virtual/net/xe9.968/queues/rx-0 (queues)
UDEV  [538.814202] add      /devices/virtual/net/xe9.968/queues/tx-0 (queues)
UDEV  [538.814664] add      /devices/virtual/net/xe9.976/queues/rx-0 (queues)
UDEV  [538.815270] add      /devices/virtual/net/xe9.976/queues/tx-0 (queues)
UDEV  [539.581034] add      /devices/virtual/net/xe9.953 (net)
UDEV  [539.634047] add      /devices/virtual/net/xe9.978 (net)
...
...
UDEV  [1277.422066] add      /devices/virtual/net/xe9.7999/queues/rx-0 (queues)
UDEV  [1277.425117] add      /devices/virtual/net/xe9.7987 (net)
UDEV  [1277.425660] add      /devices/virtual/net/xe9.7987/queues/tx-0 (queues)
UDEV  [1277.426005] add      /devices/virtual/net/xe9.7987/queues/rx-0 (queues)
UDEV  [1277.436997] add      /devices/virtual/net/xe9.7996 (net)
UDEV  [1277.437450] add      /devices/virtual/net/xe9.7996/queues/rx-0 (queues)
UDEV  [1277.437550] add      /devices/virtual/net/xe9.7996/queues/tx-0 (queues)
UDEV  [1277.473557] add      /devices/virtual/net/xe9.7997 (net)
UDEV  [1277.474560] add      /devices/virtual/net/xe9.7997/queues/tx-0 (queues)
UDEV  [1277.474615] add      /devices/virtual/net/xe9.7997/queues/rx-0 (queues)


root@NetwAA:~#
root@NetwAA:~#
root@NetwAA:~# pgrep -l ifquery
root@NetwAA:~# pgrep -l systemd
1 systemd
1925 systemd-journal
1946 systemd-udevd
2130 systemd-logind
root@NetwAA:~#

what is happening in the background in kernel ?
why is the bottle-neck ? is it due to rtnl semaphore access ?

Offline

Board footer

Powered by FluxBB