Actions
Bug #7234
closedntpd overload during IPsec session without HW acceleration
Status:
Closed
Priority:
Low
Assignee:
-
Category:
NTPD
Target version:
-
Start date:
02/08/2017
Due date:
% Done:
0%
Estimated time:
Plus Target Version:
Release Notes:
Affected Version:
2.3.x
Affected Architecture:
Description
During performance test 2440 I noticed quite strange behavior of ntpd. One overloads CPU core during IPsec session if HW acceleration is disabled:
kldstat Id Refs Address Size Name 1 1 0xffffffff80200000 225ede0 kernel
last pid: 42143; load averages: 3.41, 2.35, 1.15 up 0+00:07:28 09:19:23 54 processes: 2 running, 51 sleeping, 1 waiting CPU 0: 9.4% user, 0.0% nice, 20.4% system, 65.9% interrupt, 4.3% idle CPU 1: 18.4% user, 0.0% nice, 53.7% system, 0.0% interrupt, 27.8% idle Mem: 45M Active, 39M Inact, 140M Wired, 21M Buf, 3676M Free Swap: 3647M Total, 3647M Free PID USERNAME THR PRI NICE SIZE RES STATE C TIME CPU COMMAND 16307 root 2 20 0 30144K 17988K usem 1 1:28 *100.00% ntpd* 23986 root 17 21 0 217M 13796K uwait 0 3:15 80.66% charon 11 root 2 155 ki31 0K 32K RUN 1 6:20 12.16% idle 12 root 27 -72 - 0K 432K WAIT 1 2:56 2.49% intr 0 root 20 -8 - 0K 320K - 0 0:22 0.00% kernel
If HW acceleration is enabled there is not issues with ntpd, one sometimes can load CPU about 3-5% for several seconds:
kldstat Id Refs Address Size Name 1 4 0xffffffff80200000 225ede0 kernel 2 1 0xffffffff82611000 7577 aesni.ko
last pid: 99164; load averages: 5.32, 2.80, 1.22 up 3+14:47:13 09:04:56 54 processes: 3 running, 50 sleeping, 1 waiting CPU 0: 4.3% user, 0.0% nice, 15.7% system, 80.0% interrupt, 0.0% idle CPU 1: 8.2% user, 0.0% nice, 21.6% system, 70.2% interrupt, 0.0% idle Mem: 11M Active, 75M Inact, 153M Wired, 382M Buf, 3661M Free Swap: 3647M Total, 3647M Free PID USERNAME THR PRI NICE SIZE RES STATE C TIME CPU COMMAND 12 root 27 -72 - 0K 432K WAIT 1 51:30 200.00% intr 11 root 2 155 ki31 0K 32K RUN 1 171.4H 2.83% idle 15438 root 2 20 0 30144K 17988K usem 0 7:20 *0.49% ntpd* 53451 root 17 72 0 217M 14684K RUN 0 0:50 0.20% charon 0 root 20 -8 - 0K 320K - 0 8:28 0.00% kernel
Checked on 2440
uname -a FreeBSD pfSense.localdomain 10.3-RELEASE-p5 FreeBSD 10.3-RELEASE-p5 #0 7307492(RELENG_2_3_2): Tue Jul 19 13:29:35 CDT 2016 root@ce23-amd64-builder:/builder/pfsense-232/tmp/obj/builder/pfsense-232/tmp/FreeBSD-src/sys/pfSense amd64
Checked on 4860. For this device it does not matter is HW acceleration enabled or disabled, picture is same. (I have got strong suspicion there is an issue with HW acceleration on 4860)
uname -a FreeBSD pfSense.localdomain 10.3-RELEASE-p9 FreeBSD 10.3-RELEASE-p9 #1 5fc1b19(RELENG_2_3_2): Tue Sep 27 12:25:49 CDT 2016 root@factory23-amd64-builder:/builder/factory-232/tmp/obj/builder/factory-232/tmp/FreeBSD-src/sys/pfSense amd64
Updated by Kris Phillips almost 2 years ago
Testing this on pfSense Plus 22.01, I'm unable to reproduce any NTP CPU locking on a single thread testing with or without hardware accel. This can be Rejected unless someone can reproduce this.
Updated by Jim Pingle almost 2 years ago
- Status changed from New to Closed
- Assignee deleted (
Luiz Souza)
Actions