K 10 svn:author V 3 mav K 8 svn:date V 27 2013-03-01T02:59:56.492969Z K 7 svn:log V 2112 Add statistics to better diagnose and tune callout subsystem parameters. Setting kern.callout_stat sysctl dumps to console statistics for callouts scheduled at that moment alike to this: Scheduled callouts statistic snapshot: Callouts: 69 Buckets/CPU: 32768 Bucket size: 0.000976s C/Bk: med 0 avg 0.000526 max 10 Time: med 1.000000s avg 1580.594786s max 84945.674598s Prec: med 0.250000s avg 106.280672s max 5379.937129s Distribution: buckets time tcum prec pcum 0.007812s 2**3 0 0 2 2 0.015625s 2**4 0 0 2 4 0.031250s 2**5 13 13 3 7 0.062500s 2**6 2 15 15 22 0.125000s 2**7 4 19 1 23 0.250000s 2**8 1 20 13 36 0.500000s 2**9 11 31 11 47 1.000000s 2**10 8 39 1 48 2.000000s 2**11 4 43 3 51 4.000000s 2**12 3 46 4 55 8.000000s 2**13 5 51 1 56 16.000000s 2**14 0 51 2 58 32.000000s 2**15 3 54 1 59 64.000000s 2**16 4 58 2 61 128.000000s 2**17 1 59 3 64 256.000000s 2**18 0 59 2 66 512.000000s 2**19 1 60 2 68 1024.000000s 2**20 4 64 0 68 2048.000000s 2**21 2 66 0 68 4096.000000s 2**22 0 66 1 69 8192.000000s 2**23 2 68 0 69 65536.000000s 2**26 1 69 0 69 It would be very interesting to analyze such dumps from systems with heavy load of different kinds to see how well callout subsystem is tuned for them by default. END