第33讲 perfetto 分析cpu usage -凯发k8旗舰厅

本讲是android camera性能分析专题的第33讲，我们介绍perfetto 分析cpu usage。

更多资源：

perfetto cpu usage分析需要打开的config

需要打开scheduling events trace config:

data_sources: {
    config {
        name: "linux.process_stats"
        target_buffer: 1
        process_stats_config {
            scan_all_processes_on_start: true
        }
    }
}
data_sources: {
    config {
        name: "linux.ftrace"
        ftrace_config {
            ftrace_events: "sched/sched_switch"
            ftrace_events: "power/suspend_resume"
            ftrace_events: "sched/sched_wakeup"
            ftrace_events: "sched/sched_wakeup_new"
            ftrace_events: "sched/sched_waking"
            ftrace_events: "sched/sched_process_exit"
            ftrace_events: "sched/sched_process_free"
            ftrace_events: "task/task_newtask"
            ftrace_events: "task/task_rename"
            buffer_size_kb: 2048
            drain_period_ms: 250
        }
    }
}

注意：在对比cpu usage时，要将cpu frequency固定住来对比才有意义。

perfetto统计整个系统cpu usage

在cpu区域圈2秒：

cpu usage计算

• 该阶段总的cpu运行时间 / duration

• 12367ms/1953ms = 6.3323 = 633.23%

• 注：也可以通过top命令来看整个系统的cpu usage

perfetto统计某个进程cpu usage

perfetto统计某个进程cpu usage from sql

sql语句统计进程cpu time:

select process.name, (sum(dur)/1e9) as cpu_sec
from sched
join thread using(utid)
join process using(upid)
where ts <= (select ts 5000000000 from sched order by ts limit 1)
group by upid
order by cpu_sec desc
limit 100;

结果：

perfetto统计某个进程在各cpu上的cpu usage

有时候我们需要知道该进程在大核/小核上的cpu占比，可以使用下面的sql语句，查看cameraserver在各cpu上的cpu usage情况：

select
  process.name as process,
  cpu,
  sum(dur) / 1e9 as cpu_sec
from sched
inner join thread using(utid)
inner join process using(upid)
where process.name like "�meraserver%"
group by upid, cpu
order by cpu_sec desc
limit 300;

输出：

perfetto统计某个线程cpu usage

perfetto统计某个线程在各cpu上的cpu usage from sql

有时候我们需要知道该线程在大核/小核上的cpu占比，可以使用下面的sql语句，查看c3dev-0-reqqueu在各cpu上的cpu usage情况：

select
  thread.name as thread,
  cpu,
  sum(dur) / 1e9 as cpu_sec
from sched
inner join thread using(utid)
inner join process using(upid)
where thread.name like "�dev-0-reqqueu%"
group by utid, cpu
order by cpu_sec desc
limit 300;

输出: