diff --git a/README.md b/README.md new file mode 100644 index 0000000..86225f3 --- /dev/null +++ b/README.md @@ -0,0 +1,75 @@ +# pmc-cloud-tools + +Linux tools for measuring PMCs (Performance Monitoring Counters) in the cloud. + +## Architectural PMCs + +These are seven PMCs that are enabled on some newer AWS EC2 types, eg, m4.16xl. See The PMCs of EC2. + +### pmcarch + +Summarizes the architectural PMCs. Eg: + +
+# ./pmcarch +K_CYCLES K_INSTR IPC BR_RETIRED BR_MISPRED BMR% LLCREF LLCMISS LLC% +96163187 87166313 0.91 19730994925 679187299 3.44 656597454 174313799 73.45 +93988372 87205023 0.93 19669256586 724072315 3.68 666041693 169603955 74.54 +93863787 86981089 0.93 19548779510 669172769 3.42 649844207 176100680 72.90 +93739565 86349653 0.92 19339320671 634063527 3.28 642506778 181385553 71.77 +94495981 86815232 0.92 19482504710 648954409 3.33 628548666 180975066 71.21 ++ +USAGE: + +
+# ./pmcarch -h
+USAGE: pmcarch {-C CPU | -p PID | -c CMD} [interval [duration]]
+ -C CPU # measure this CPU only
+ -p PID # measure this PID only
+ -c 'CMD' # measure this command only (quote it)
+ interval # output interval in secs (default 1)
+ duration # total seconds (default infinityish)
+ eg,
+ pmcarch # show stats across all CPUs
+ pmcarch 5 # show stats every 5 seconds
+ pmcarch -C 0 # measure CPU 0 only
+ pmcarch -p 181 # measure PID 181 only
+ pmcarch -c 'cksum /boot/*' # measure run and measure this cmd
+
+
+## All PMCs
+
+These are available on some AWS Nitro hypervisor instance types. Eg, c5.9xl.
+
+### tlbstat
+
+Summarizes TLB (Translation Lookaside Buffer) statistics (an MMU cache). Eg:
+
++# ./tlbstat +K_CYCLES K_INSTR IPC DTLB_WALKS ITLB_WALKS K_DTLBCYC K_ITLBCYC DTLB% ITLB% +93091508 86971957 0.93 135562325 45028565 3053416 1121015 3.28 1.20 +94354781 88708445 0.94 136898873 49736338 3135383 1199061 3.32 1.27 +94274360 88372901 0.94 138668282 48503863 3200281 1194388 3.39 1.27 +92040379 86763153 0.94 133141376 44859310 3060742 1118921 3.33 1.22 +92152495 87144845 0.95 135446984 50308740 3156780 1200598 3.43 1.30 ++ +USAGE: + +
+# ./tlbstat -h
+USAGE: tlbstat {-C CPU | -p PID | -c CMD} [interval [duration]]
+ -C CPU # measure this CPU only
+ -p PID # measure this PID only
+ -c 'CMD' # measure this command only (quote it)
+ interval # output interval in secs (default 1)
+ duration # total seconds (default infinityish)
+ eg,
+ tlbstat # show stats across all CPUs
+ tlbstat 5 # show stats every 5 seconds
+ tlbstat -C 0 # measure CPU 0 only
+ tlbstat -p 181 # measure PID 181 only
+ tlbstat -c 'cksum /boot/*' # measure run and measure this cmd
+
diff --git a/pmcarch b/pmcarch
index df18cc2..c070c66 100755
--- a/pmcarch
+++ b/pmcarch
@@ -2,9 +2,13 @@
#
# pmcarch - measure and summarize architectural PMCs.
#
+# USAGE: pmcarch {-C CPU | -p PID | -c CMD} [interval [duration]]
+#
# PMCs are performance monitoring counters. This script uses the architectural
# set to report on:
#
+# - K_CYCLES: CPU Cycles x 1000
+# - K_INSTR: CPU Instructions x 1000
# - IPC: instructions-per-cycle
# - BMR%: branch misprediction ration, as a percentage
# - LLC%: last level cache hit ratio, as a percentage
@@ -14,16 +18,62 @@
#
# 07-Apr-2017 Brendan Gregg Created this.
+function usage {
+ cat <<-END >&2
+ USAGE: pmcarch {-C CPU | -p PID | -c CMD} [interval [duration]]
+ -C CPU # measure this CPU only
+ -p PID # measure this PID only
+ -c 'CMD' # measure this command only (quote it)
+ interval # output interval in secs (default 1)
+ duration # total seconds (default infinityish)
+ eg,
+ pmcarch # show stats across all CPUs
+ pmcarch 5 # show stats every 5 seconds
+ pmcarch -C 0 # measure CPU 0 only
+ pmcarch -p 181 # measure PID 181 only
+ pmcarch -c 'cksum /boot/*' # measure run and measure this cmd
+END
+ exit
+}
+
+opt_cpu=0; opt_pid=0; opt_cmd=0; cpu=""; pid=""; cmd=""
+
+while getopts C:p:c:h opt
+do
+ case $opt in
+ C) opt_cpu=1; cpu=$OPTARG ;;
+ p) opt_pid=1; pid=$OPTARG ;;
+ c) opt_cmd=1; cmd=$OPTARG ;;
+ h|?) usage ;;
+ esac
+done
+shift $(( $OPTIND - 1 ))
+
+if (( opt_cpu + opt_pid + opt_cmd > 1 )); then
+ echo >&2 "ERROR: pick one of -C, -p, -c"
+ usage
+fi
secs=${1-1} # default 1 second
duration=${2-999999999} # default semi-infinite seconds
hlines=25 # lines to repeat header
+target=-a
+(( opt_cpu )) && target="-C $cpu sleep $duration"
+(( opt_pid )) && target="-p $pid sleep $duration"
+(( opt_cmd )) && target="$cmd"
+
+if (( opt_pid )); then
+ if [ ! -d /proc/$pid ]; then
+ echo >&2 "ERROR: Can't find PID $pid. Exiting."
+ exit
+ fi
+fi
# order of PMCs attempts to workaround a 3.13 issue
# note that instructions is last on purpose, it triggers output
-perf stat -e cycles \
+perf stat -e cycles -e cycles \
-e r4f2e -e r412e \
-e r00c4 -e r00c5 -e instructions \
- -a -I $(( secs * 1000 )) sleep $duration 2>&1 | awk -v hlines=$hlines '
+ -I $(( secs * 1000 )) $target 2>&1 | awk -v hlines=$hlines '
BEGIN {
htxt = sprintf("%-10s %-10s %5s %-12s %-10s %5s %-11s %-10s %5s",
"K_CYCLES", "K_INSTR", "IPC", "BR_RETIRED", "BR_MISPRED", "BMR%",
@@ -51,6 +101,7 @@ perf stat -e cycles \
}
ins = $2
if (llcref == 0) { llcref = 1; llcmiss = 0 }
+ if (cycles == 0) { cycles = 1 } # PMCs are broken, or no events
printf("%-10d %-10d %5.2f %-12d %-10d %5.2f %-11d %-10d %5.2f\n",
cycles / 1000, ins / 1000, ins / cycles, brretired, brmiss,
diff --git a/tlbstat b/tlbstat
index acc6700..670be7e 100755
--- a/tlbstat
+++ b/tlbstat
@@ -5,7 +5,7 @@
# TLB: Translation Lookaside Buffer (the MMU page translation cache)
# PMC: Performance Monitoring Counters
#
-# USAGE: tlbstat [interval [duration]]
+# USAGE: tlbstat {-C CPU | -p PID | -c CMD} [interval [duration]]
#
# REQUIREMENTS: perf command, TLB PMCs (counters listed below)
#
@@ -26,12 +26,58 @@
#
# 08-Jan-2018 Brendan Gregg Created this.
+function usage {
+ cat <<-END >&2
+ USAGE: tlbstat {-C CPU | -p PID | -c CMD} [interval [duration]]
+ -C CPU # measure this CPU only
+ -p PID # measure this PID only
+ -c 'CMD' # measure this command only (quote it)
+ interval # output interval in secs (default 1)
+ duration # total seconds (default infinityish)
+ eg,
+ tlbstat # show stats across all CPUs
+ tlbstat 5 # show stats every 5 seconds
+ tlbstat -C 0 # measure CPU 0 only
+ tlbstat -p 181 # measure PID 181 only
+ tlbstat -c 'cksum /boot/*' # measure run and measure this cmd
+END
+ exit
+}
+
+opt_cpu=0; opt_pid=0; opt_cmd=0; cpu=""; pid=""; cmd=""
+
+while getopts C:p:c:h opt
+do
+ case $opt in
+ C) opt_cpu=1; cpu=$OPTARG ;;
+ p) opt_pid=1; pid=$OPTARG ;;
+ c) opt_cmd=1; cmd=$OPTARG ;;
+ h|?) usage ;;
+ esac
+done
+shift $(( $OPTIND - 1 ))
+
+if (( opt_cpu + opt_pid + opt_cmd > 1 )); then
+ echo >&2 "ERROR: pick one of -C, -p, -c"
+ usage
+fi
secs=${1-1} # default 1 second
duration=${2-999999999} # default semi-infinite seconds
hlines=25 # lines to repeat header
+target=-a
+(( opt_cpu )) && target="-C $cpu sleep $duration"
+(( opt_pid )) && target="-p $pid sleep $duration"
+(( opt_cmd )) && target="$cmd"
+
+if (( opt_pid )); then
+ if [ ! -d /proc/$pid ]; then
+ echo >&2 "ERROR: Can't find PID $pid. Exiting."
+ exit
+ fi
+fi
# note that instructions is last on purpose, it triggers output
-perf stat -e cycles \
+perf stat -e cycles -e cycles \
-e dtlb_load_misses.miss_causes_a_walk \
-e dtlb_store_misses.miss_causes_a_walk \
-e itlb_misses.miss_causes_a_walk \
@@ -39,7 +85,7 @@ perf stat -e cycles \
-e dtlb_store_misses.walk_active \
-e itlb_misses.walk_active \
-e instructions \
- -a -I $(( secs * 1000 )) sleep $duration 2>&1 | awk -v hlines=$hlines '
+ -I $(( secs * 1000 )) $target 2>&1 | awk -v hlines=$hlines '
BEGIN {
htxt = sprintf("%-10s %-10s %5s %-10s %-10s %-10s %-10s %5s %5s",
"K_CYCLES", "K_INSTR", "IPC", "DTLB_WALKS", "ITLB_WALKS",
@@ -63,7 +109,7 @@ perf stat -e cycles \
header = hlines
}
ins = $2
- if (cycles == 0) { cycles = 1 } # and PMCs are broken
+ if (cycles == 0) { cycles = 1 } # PMCs are broken, or no events
printf("%-10d %-10d %5.2f %-10d %-10d %-10d %-10d %5.2f %5.2f\n",
cycles / 1000, ins / 1000, ins / cycles,