[root@localhost ~]# perf stat -da
Performance counter stats for 'system wide':
11822.538041 cpu-clock (msec) # 2.000 CPUs utilized
1,006 context-switches # 0.085 K/sec
12 cpu-migrations # 0.001 K/sec
1,113 page-faults # 0.094 K/sec
<not supported> cycles
<not supported> instructions
<not supported> branches
<not supported> branch-misses
<not supported> L1-dcache-loads
<not supported> L1-dcache-load-misses
<not supported> LLC-loads
<not supported> LLC-load-misses
[root@localhost ~]# perf stat -da
Performance counter stats for 'system wide':
16614.742348 cpu-clock (msec) # 2.001 CPUs utilized
2,537 context-switches # 0.153 K/sec
49 cpu-migrations # 0.003 K/sec
926 page-faults # 0.056 K/sec
693,283 cycles # 0.000 GHz (66.71%)
0 instructions # 0.00 insn per cycle (83.38%)
45,732,264 branches # 2.753 M/sec (83.31%)
713,155 branch-misses # 1.56% of all branches (83.34%)
0 L1-dcache-loads # 0.000 K/sec (83.37%)
2,784,870 L1-dcache-load-misses # 0.00% of all L1-dcache hits (83.29%)
<not supported> LLC-loads
<not supported> LLC-load-misses
From: Alexey Pirogov Sent: Sunday, 11 December 2016 17:04 To: mechanical-sympathy Reply To: mechanica...@googlegroups.com Subject: Profiling using hardware counters on Mac OS X |
Hi Alexey,As far as I see, there are few options to try:1. Try to use BOOTCAMP and install windows on your laptop. Vtune should be available in this environment (I used it on HSW laptops).2. I'm not sure if dtruss supports PMU (hardware counters), but if so, it may be possible to write a wrapper over dtrace and pass it to JMH via JMH_PERF environment variable. JMH will run your wrapper and parse it's output. The wrapper will run Dtruss and convert it's output to a perf format.3. Try to find perf/oprofile in macports or homebrew software repos.Anyway, I'm not sure it's a good idea to use PMU in virtualized environment.BTW, performance analysis on so low level may require "clean" system (pc/mac) without any background "noise" and disabled performance management/updates/... .--Sergey
Sent from my BlackBerry
From: Alexey PirogovSent: Sunday, 11 December 2016 17:04To: mechanical-sympathy
Reply To: mechanical-sympathy@googlegroups.com
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsub...@googlegroups.com.
--
You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsub...@googlegroups.com.
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsubscribe...@googlegroups.com.
--
You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsubscribe...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
From: Avi Kivity Sent: Monday, 12 December 2016 00:05 Reply To: mechanica...@googlegroups.com Subject: Re: Profiling using hardware counters on Mac OS X |
AWS partitions machines rather than shares them, so interaction
on the cpu core is limited. L3 caches are shared, but if you get
a large instance (largest = full machine, next largest = full
socket) you avoid that too.
--