1*4882a593Smuzhiyunperf-timechart(1) 2*4882a593Smuzhiyun================= 3*4882a593Smuzhiyun 4*4882a593SmuzhiyunNAME 5*4882a593Smuzhiyun---- 6*4882a593Smuzhiyunperf-timechart - Tool to visualize total system behavior during a workload 7*4882a593Smuzhiyun 8*4882a593SmuzhiyunSYNOPSIS 9*4882a593Smuzhiyun-------- 10*4882a593Smuzhiyun[verse] 11*4882a593Smuzhiyun'perf timechart' [<timechart options>] {record} [<record options>] 12*4882a593Smuzhiyun 13*4882a593SmuzhiyunDESCRIPTION 14*4882a593Smuzhiyun----------- 15*4882a593SmuzhiyunThere are two variants of perf timechart: 16*4882a593Smuzhiyun 17*4882a593Smuzhiyun 'perf timechart record <command>' to record the system level events 18*4882a593Smuzhiyun of an arbitrary workload. By default timechart records only scheduler 19*4882a593Smuzhiyun and CPU events (task switches, running times, CPU power states, etc), 20*4882a593Smuzhiyun but it's possible to record IO (disk, network) activity using -I argument. 21*4882a593Smuzhiyun 22*4882a593Smuzhiyun 'perf timechart' to turn a trace into a Scalable Vector Graphics file, 23*4882a593Smuzhiyun that can be viewed with popular SVG viewers such as 'Inkscape'. Depending 24*4882a593Smuzhiyun on the events in the perf.data file, timechart will contain scheduler/cpu 25*4882a593Smuzhiyun events or IO events. 26*4882a593Smuzhiyun 27*4882a593Smuzhiyun In IO mode, every bar has two charts: upper and lower. 28*4882a593Smuzhiyun Upper bar shows incoming events (disk reads, ingress network packets). 29*4882a593Smuzhiyun Lower bar shows outgoing events (disk writes, egress network packets). 30*4882a593Smuzhiyun There are also poll bars which show how much time application spent 31*4882a593Smuzhiyun in poll/epoll/select syscalls. 32*4882a593Smuzhiyun 33*4882a593SmuzhiyunTIMECHART OPTIONS 34*4882a593Smuzhiyun----------------- 35*4882a593Smuzhiyun-o:: 36*4882a593Smuzhiyun--output=:: 37*4882a593Smuzhiyun Select the output file (default: output.svg) 38*4882a593Smuzhiyun-i:: 39*4882a593Smuzhiyun--input=:: 40*4882a593Smuzhiyun Select the input file (default: perf.data unless stdin is a fifo) 41*4882a593Smuzhiyun-w:: 42*4882a593Smuzhiyun--width=:: 43*4882a593Smuzhiyun Select the width of the SVG file (default: 1000) 44*4882a593Smuzhiyun-P:: 45*4882a593Smuzhiyun--power-only:: 46*4882a593Smuzhiyun Only output the CPU power section of the diagram 47*4882a593Smuzhiyun-T:: 48*4882a593Smuzhiyun--tasks-only:: 49*4882a593Smuzhiyun Don't output processor state transitions 50*4882a593Smuzhiyun-p:: 51*4882a593Smuzhiyun--process:: 52*4882a593Smuzhiyun Select the processes to display, by name or PID 53*4882a593Smuzhiyun-f:: 54*4882a593Smuzhiyun--force:: 55*4882a593Smuzhiyun Don't complain, do it. 56*4882a593Smuzhiyun--symfs=<directory>:: 57*4882a593Smuzhiyun Look for files with symbols relative to this directory. 58*4882a593Smuzhiyun-n:: 59*4882a593Smuzhiyun--proc-num:: 60*4882a593Smuzhiyun Print task info for at least given number of tasks. 61*4882a593Smuzhiyun-t:: 62*4882a593Smuzhiyun--topology:: 63*4882a593Smuzhiyun Sort CPUs according to topology. 64*4882a593Smuzhiyun--highlight=<duration_nsecs|task_name>:: 65*4882a593Smuzhiyun Highlight tasks (using different color) that run more than given 66*4882a593Smuzhiyun duration or tasks with given name. If number is given it's interpreted 67*4882a593Smuzhiyun as number of nanoseconds. If non-numeric string is given it's 68*4882a593Smuzhiyun interpreted as task name. 69*4882a593Smuzhiyun--io-skip-eagain:: 70*4882a593Smuzhiyun Don't draw EAGAIN IO events. 71*4882a593Smuzhiyun--io-min-time=<nsecs>:: 72*4882a593Smuzhiyun Draw small events as if they lasted min-time. Useful when you need 73*4882a593Smuzhiyun to see very small and fast IO. It's possible to specify ms or us 74*4882a593Smuzhiyun suffix to specify time in milliseconds or microseconds. 75*4882a593Smuzhiyun Default value is 1ms. 76*4882a593Smuzhiyun--io-merge-dist=<nsecs>:: 77*4882a593Smuzhiyun Merge events that are merge-dist nanoseconds apart. 78*4882a593Smuzhiyun Reduces number of figures on the SVG and makes it more render-friendly. 79*4882a593Smuzhiyun It's possible to specify ms or us suffix to specify time in 80*4882a593Smuzhiyun milliseconds or microseconds. 81*4882a593Smuzhiyun Default value is 1us. 82*4882a593Smuzhiyun 83*4882a593SmuzhiyunRECORD OPTIONS 84*4882a593Smuzhiyun-------------- 85*4882a593Smuzhiyun-P:: 86*4882a593Smuzhiyun--power-only:: 87*4882a593Smuzhiyun Record only power-related events 88*4882a593Smuzhiyun-T:: 89*4882a593Smuzhiyun--tasks-only:: 90*4882a593Smuzhiyun Record only tasks-related events 91*4882a593Smuzhiyun-I:: 92*4882a593Smuzhiyun--io-only:: 93*4882a593Smuzhiyun Record only io-related events 94*4882a593Smuzhiyun-g:: 95*4882a593Smuzhiyun--callchain:: 96*4882a593Smuzhiyun Do call-graph (stack chain/backtrace) recording 97*4882a593Smuzhiyun 98*4882a593SmuzhiyunEXAMPLES 99*4882a593Smuzhiyun-------- 100*4882a593Smuzhiyun 101*4882a593Smuzhiyun$ perf timechart record git pull 102*4882a593Smuzhiyun 103*4882a593Smuzhiyun [ perf record: Woken up 13 times to write data ] 104*4882a593Smuzhiyun [ perf record: Captured and wrote 4.253 MB perf.data (~185801 samples) ] 105*4882a593Smuzhiyun 106*4882a593Smuzhiyun$ perf timechart 107*4882a593Smuzhiyun 108*4882a593Smuzhiyun Written 10.2 seconds of trace to output.svg. 109*4882a593Smuzhiyun 110*4882a593SmuzhiyunRecord system-wide timechart: 111*4882a593Smuzhiyun 112*4882a593Smuzhiyun $ perf timechart record 113*4882a593Smuzhiyun 114*4882a593Smuzhiyun then generate timechart and highlight 'gcc' tasks: 115*4882a593Smuzhiyun 116*4882a593Smuzhiyun $ perf timechart --highlight gcc 117*4882a593Smuzhiyun 118*4882a593SmuzhiyunRecord system-wide IO events: 119*4882a593Smuzhiyun 120*4882a593Smuzhiyun $ perf timechart record -I 121*4882a593Smuzhiyun 122*4882a593Smuzhiyun then generate timechart: 123*4882a593Smuzhiyun 124*4882a593Smuzhiyun $ perf timechart 125*4882a593Smuzhiyun 126*4882a593SmuzhiyunSEE ALSO 127*4882a593Smuzhiyun-------- 128*4882a593Smuzhiyunlinkperf:perf-record[1] 129
