Device polling / possible bug?

xibalba

hi
with kern.polling.enable=1 i expieriece the following.

TOP:
last pid: 36937; load averages: 0.90, 0.92, 0.97 up 0+05:54:06 07:57:46
48 processes: 2 running, 45 sleeping, 1 stopped
CPU states: 0.0% user, 0.0% nice, 95.7% system, 4.3% interrupt, 0.0% idle
Mem: 53M Active, 26M Inact, 28M Wired, 34M Buf, 135M Free
Swap: 1024M Total, 1024M Free

systat -vm output (i sftp'd it over):
notice interupt level
3 users Load 0.96 0.93 0.97 Oct 21 07:58

Mem:KB REAL VIRTUAL VN PAGER SWAP PAGER
Tot Share Tot Share Free in out in out
Act 62780 5448 178884 7196 138028 count
All 108144 7800 13997824 10032 pages
Interrupts
Proc:r p d s w Csw Trp Sys Int Sof Flt 5 cow 1129 total
1 5 41 312k 110 335 10353 1 17 28228 wire 1001 0: clk
54272 act 6: fdc0
96.1%Sys 3.9%Intr 0.0%User 0.0%Nice 0.0%Idl 26280 inact 7: ppc0
| | | | | | | | | | cache 128 8: rtc
================================================ 138028 free 10: dc0
daefr 11: xl0
Namei Name-cache Dir-cache 10 prcfr 14: ata
Calls hits % hits % react 15: ata
77 77 100 pdwake
9 zfod pdpgs
Disks ad0 9 ozfod intrn
KB/t 16.00 %slo-z 35136 buf
tps 0 17 tfree 10 dirtybuf
MB/s 0.00 17444 desiredvnodes
% busy 0 2748 numvnodes
252 freevnodes

when i disable kern.polling.enable=0
this is the output of top just a few seconds later

last pid: 37120; load averages: 0.86, 0.93, 0.96 up 0+05:57:31 08:01:11
48 processes: 2 running, 45 sleeping, 1 stopped
CPU states: 0.4% user, 0.0% nice, 0.4% system, 0.8% interrupt, 98.4% idle
Mem: 53M Active, 26M Inact, 28M Wired, 34M Buf, 135M Free
Swap: 1024M Total, 1024M Free

systat output
3 users Load 0.63 0.87 0.94 Oct 21 08:01

Mem:KB REAL VIRTUAL VN PAGER SWAP PAGER
Tot Share Tot Share Free in out in out
Act 62784 5448 178884 7196 137940 count
All 108236 7800 13997984 10032 pages
Interrupts
Proc:r p d s w Csw Trp Sys Int Sof Flt 5 cow 1176 total
5 42 712 110 333 1436 1 17 28268 wire 1000 0: clk
54280 act 6: fdc0
0.3%Sys 1.1%Intr 0.2%User 0.2%Nice 98.3%Idl 26324 inact 7: ppc0
| | | | | | | | | | cache 128 8: rtc
+ 137940 free 1 10: dc0
daefr 47 11: xl0
Namei Name-cache Dir-cache 10 prcfr 14: ata
Calls hits % hits % react 15: ata
87 86 99 pdwake
9 zfod pdpgs
Disks ad0 9 ozfod intrn
KB/t 9.00 %slo-z 35136 buf
tps 0 17 tfree 6 dirtybuf
MB/s 0.00 17444 desiredvnodes
% busy 0 2760 numvnodes
252 freevnodes

the system % in top and the amount of interrupts went down significantly.
i think maybe device polling just should be done on shitty NICs.
i have an xl and dc based network cards.

billm

Cute. Did you set the sysctl manually? I ask because I think that sysctl is supposed to be deprecated in favor of ifconfig polling/ifconfig -polling. A quick glance at the HEAD branch shows that we're using ifconfig, I didn't look over at RELENG_1, so I'm not sure what release uses offhand.

–Bill

xibalba

i used sysctl to disable device polling, but then i just switched it off via the GUI.