-问题起因
近期线上一组服务中，个别节点服务器CPU使用率很低，只有其他1/4。排除业务不均，曾怀疑是系统top统计错误，从Erlang调度器的利用率调查 找到通过erlang:statistics(scheduler_wall_time) 查看服务器CPU低的机器调度器实际的CPU利用率很高接近100%，而其他机器都不到30%。
分析不同业务服务，发现只有在node 中进程数采用调度器CPU利用低这个问题。
-Whatsapp 案例
erlang方面能找到案例不多，幸运的发现whatsapp 给出了类似案例详细的分析：
First bottleneck showed up at 425K. System ran into a lot of contention. Work stopped.
Instrumented the scheduler to measure how much useful work is being done, or sleeping, or spinning.
Under load it started to hit sleeping locks so 35-45% CPU was being used across the system but the schedulers are at 95% utilization.
whatsapp 在单机425K连接时遇到瓶颈，虚机实际只35~45%cpu却给系统造成了95%的cpu。
文中没有提到具体细节，关于scheduler：
1. +swt low Set the scheduler wake up threshold to low because schedulers would go to sleep and would never wake up
2. 设置进程优先级为实时 Run BEAM at real-time priority so that other things like cron jobs don’t interrupt schedule. Prevents glitches that would cause backlogs of important user traffic
3. 禁用spin（ patch beam）Patch to dial down spin counts so the scheduler wouldn’t spin，+ssct 1 (via patch; scheduler spin count)
-工具分析
通过微博私信，请教了下郑思瑶，推荐VTune分析，推测是大量进程时调度器消耗过大。
通过Intel 官方网站，填写注册信息，会立即回复邮件下载地址，并给30天试用期。
速度很慢，建议挂着VPN下载；VTune 的linux版本命令行模式使用很简单：
tar -zxf vtune_amplifier_xe_2015.tar.gz
cd vtune_amplifier_xe_2015
./install.sh
cd /opt/intel/vtune_amplifier_xe_2015.1.0.367959/
source amplxe-vars.sh
amplxe-cl -collect lightweight-hotspots -run-pass-thru=--no-altstack -target-pid=1575
amplxe-cl -report hotspots
可直接线上执行，不影响服务正常运行，得到如下结果：
Summary ------- Elapsed Time: 19.345 CPU Time: 182.023 Average CPU Usage: 9.155 CPI Rate: 1.501 Function Module CPU Time:Self ------------------------------------------- ------------------ ------------- sched_spin_wait beam.smp 72.754 raw_local_irq_enable vmlinux 19.282 process_main beam.smp 10.476 ethr_native_atomic32_read beam.smp 8.337 [email protected]0xffffffff8100af60 vmlinux 3.007 __pthread_mutex_lock libpthread-2.12.so 2.342 raw_local_irq_restore vmlinux 1.973 __sched_yield libc-2.12.so 1.913 pthread_mutex_unlock libpthread-2.12.so 1.553 __audit_syscall_exit vmlinux 1.192 system_call vmlinux 1.156 erts_thr_yield beam.smp 1.114 handle_delayed_dealloc beam.smp 0.977 update beam.smp 0.828 raw_local_irq_enable vmlinux 0.780
可以看到sched_spin_wait占用了 40% 的CPU时间
#define ERTS_SCHED_SPIN_UNTIL_YIELD 100
2121 static erts_aint32_t
2122 sched_spin_wait(ErtsSchedulerSleepInfo *ssi, int spincount)
2123 {
2124 int until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD;
2125 int sc = spincount;
2126 erts_aint32_t flgs;
2127
2128 do {
2129 flgs = erts_smp_atomic32_read_acqb(&ssi->flags);
2130 if ((flgs & (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING))
2131 != (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) {
2132 break;
2133 }
2134 ERTS_SPIN_BODY;
2135 if (--until_yield == 0) {
2136 until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD;
2137 erts_thr_yield();
2138 }
2139 } while (--sc > 0);
2140 return flgs;
2141 }
默认spincount = 10000，但每次都有atom 读操作，原子操作一般几十到几百CPU周期，导致这个忙等待 实际执行完的话会很长。
同时也找到对应的配置：
+sbwt none|very_short|short|medium|long|very_longSet scheduler busy wait threshold. Default is medium. The threshold determines how long schedulers should busy wait when running out of work before going to sleep.
启动参数：+sbwt none 即可见spin彻底关掉，而不必像whatsapp patch beam解决。