Skip to contentSkip to conversation

perf test stat_bpf_counter.sh: Stabilize the test results

PATCH Applied [email protected] 2 participants 3 messages Jun 25, 2024 – Jun 27, 2024 perf events

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 perf test stat_bpf_counter.sh: Stabilize the test results

From: Veronika Molnarova <[email protected]>

The test has been failing for some time when two separate runs of perf benchmarks are recorded for cycles events and their counts are compared, while once the recording was done with option --bpf-counters and once without it. It is expected that the count of the samples should be within a certain range, firstly the difference was set to be within 10%, which was then later raised to 20%. However, the test case keeps failing on certain architectures as recording the provided benchmark can produce completely different counts based on the current load of the system.

Sampling two separate runs on intel-eaglestream-spr-13 of "perf stat --no-big-num -e cycles -- perf bench sched messaging -g 1 -l 100 -t":

Performance counter stats for 'perf bench sched messaging -g 1 -l 100 -t':

         396782898      cycles

       0.010051983 seconds time elapsed

       0.008664000 seconds user
       0.097058000 seconds sys

Performance counter stats for 'perf bench sched messaging -g 1 -l 100 -t':

        1431133032      cycles

       0.021803714 seconds time elapsed

       0.023377000 seconds user
       0.349918000 seconds sys

, which is ranging from 400mil to 1400mil samples.

Instead of recording the cycles use instructions event, which provides more stable values. At the same time change the tested workload to one of the provided testing workloads by perf that is not based on a scheduler, which can provide another dependency on the current load.

Sampling instructions event with the new workload provide much more stable results on intel-eaglestream-spr-13 of "perf stat --no-big-num -e instructions -- perf test -w brstack":

Performance counter stats for 'perf test -w brstack':

          64584494      instructions

       0.009173945 seconds time elapsed

       0.007262000 seconds user
       0.002071000 seconds sys

Performance counter stats for 'perf test -w brstack':

          64672669      instructions

       0.008888135 seconds time elapsed

       0.005018000 seconds user
       0.004018000 seconds sys
---
 tools/perf/tests/shell/stat_bpf_counters.sh | 36 ++++++++++-----------
 1 file changed, 18 insertions(+), 18 deletions(-)
tools/perf/tests/shell/stat_bpf_counters.sh +18 -18
index 61f8149d854e..f250b7d6f773 100755
@@ -4,7 +4,7 @@
 
 set -e
 
-workload="perf bench sched messaging -g 1 -l 100 -t"
+workload="perf test -w brstack"
 
 # check whether $2 is within +/- 20% of $1
 compare_number()
@@ -25,15 +25,15 @@ compare_number()
 
 check_counts()
 {
-	base_cycles=$1
-	bpf_cycles=$2
+	base_instructions=$1
+	bpf_instructions=$2
 
-	if [ "$base_cycles" = "<not" ]; then
-		echo "Skipping: cycles event not counted"
+	if [ "$base_instructions" = "<not" ]; then
+		echo "Skipping: instructions event not counted"
 		exit 2
 	fi
-	if [ "$bpf_cycles" = "<not" ]; then
-		echo "Failed: cycles not counted with --bpf-counters"
+	if [ "$bpf_instructions" = "<not" ]; then
+		echo "Failed: instructions not counted with --bpf-counters"
 		exit 1
 	fi
 }
@@ -41,29 +41,29 @@ check_counts()
 test_bpf_counters()
 {
 	printf "Testing --bpf-counters "
-	base_cycles=$(perf stat --no-big-num -e cycles -- $workload 2>&1 | awk '/cycles/ {print $1}')
-	bpf_cycles=$(perf stat --no-big-num --bpf-counters -e cycles -- $workload  2>&1 | awk '/cycles/ {print $1}')
-	check_counts $base_cycles $bpf_cycles
-	compare_number $base_cycles $bpf_cycles
+	base_instructions=$(perf stat --no-big-num -e instructions -- $workload 2>&1 | awk '/instructions/ {print $1}')
+	bpf_instructions=$(perf stat --no-big-num --bpf-counters -e instructions -- $workload  2>&1 | awk '/instructions/ {print $1}')
+	check_counts $base_instructions $bpf_instructions
+	compare_number $base_instructions $bpf_instructions
 	echo "[Success]"
 }
 
 test_bpf_modifier()
 {
 	printf "Testing bpf event modifier "
-	stat_output=$(perf stat --no-big-num -e cycles/name=base_cycles/,cycles/name=bpf_cycles/b -- $workload 2>&1)
-	base_cycles=$(echo "$stat_output"| awk '/base_cycles/ {print $1}')
-	bpf_cycles=$(echo "$stat_output"| awk '/bpf_cycles/ {print $1}')
-	check_counts $base_cycles $bpf_cycles
-	compare_number $base_cycles $bpf_cycles
+	stat_output=$(perf stat --no-big-num -e instructions/name=base_instructions/,instructions/name=bpf_instructions/b -- $workload 2>&1)
+	base_instructions=$(echo "$stat_output"| awk '/base_instructions/ {print $1}')
+	bpf_instructions=$(echo "$stat_output"| awk '/bpf_instructions/ {print $1}')
+	check_counts $base_instructions $bpf_instructions
+	compare_number $base_instructions $bpf_instructions
 	echo "[Success]"
 }
 
 # skip if --bpf-counters is not supported
-if ! perf stat -e cycles --bpf-counters true > /dev/null 2>&1; then
+if ! perf stat -e instructions --bpf-counters true > /dev/null 2>&1; then
 	if [ "$1" = "-v" ]; then
 		echo "Skipping: --bpf-counters not supported"
-		perf --no-pager stat -e cycles --bpf-counters true || true
+		perf --no-pager stat -e instructions --bpf-counters true || true
 	fi
 	exit 2
 fi
Signature · 1 line
2.43.0
maintainer Acked-by on v2 1/1
#1 lore raw

Hello,

Quoted 55 lines · On Tue, Jun 25, 2024 at 11:20:01AM +0200, [email protected] wrote:

From: Veronika Molnarova <[email protected]>

The test has been failing for some time when two separate runs of perf benchmarks are recorded for cycles events and their counts are compared, while once the recording was done with option --bpf-counters and once without it. It is expected that the count of the samples should be within a certain range, firstly the difference was set to be within 10%, which was then later raised to 20%. However, the test case keeps failing on certain architectures as recording the provided benchmark can produce completely different counts based on the current load of the system.

Sampling two separate runs on intel-eaglestream-spr-13 of "perf stat --no-big-num -e cycles -- perf bench sched messaging -g 1 -l 100 -t":

Performance counter stats for 'perf bench sched messaging -g 1 -l 100 -t':

         396782898      cycles

       0.010051983 seconds time elapsed

       0.008664000 seconds user
       0.097058000 seconds sys

Performance counter stats for 'perf bench sched messaging -g 1 -l 100 -t':

        1431133032      cycles

       0.021803714 seconds time elapsed

       0.023377000 seconds user
       0.349918000 seconds sys

, which is ranging from 400mil to 1400mil samples.

Instead of recording the cycles use instructions event, which provides more stable values. At the same time change the tested workload to one of the provided testing workloads by perf that is not based on a scheduler, which can provide another dependency on the current load.

Sampling instructions event with the new workload provide much more stable results on intel-eaglestream-spr-13 of "perf stat --no-big-num -e instructions -- perf test -w brstack":

Performance counter stats for 'perf test -w brstack':

          64584494      instructions

       0.009173945 seconds time elapsed

       0.007262000 seconds user
       0.002071000 seconds sys

Performance counter stats for 'perf test -w brstack':

Quoted 8 lines
          64672669      instructions

       0.008888135 seconds time elapsed

       0.005018000 seconds user
       0.004018000 seconds sys

Thanks,
Namhyung

Quoted 83 lines
---
 tools/perf/tests/shell/stat_bpf_counters.sh | 36 ++++++++++-----------
 1 file changed, 18 insertions(+), 18 deletions(-)
tools/perf/tests/shell/stat_bpf_counters.sh +18 -18
index 61f8149d854e..f250b7d6f773 100755
@@ -4,7 +4,7 @@
 
 set -e
 
-workload="perf bench sched messaging -g 1 -l 100 -t"
+workload="perf test -w brstack"
 
 # check whether $2 is within +/- 20% of $1
 compare_number()
@@ -25,15 +25,15 @@ compare_number()
 
 check_counts()
 {
-	base_cycles=$1
-	bpf_cycles=$2
+	base_instructions=$1
+	bpf_instructions=$2
 
-	if [ "$base_cycles" = "<not" ]; then
-		echo "Skipping: cycles event not counted"
+	if [ "$base_instructions" = "<not" ]; then
+		echo "Skipping: instructions event not counted"
 		exit 2
 	fi
-	if [ "$bpf_cycles" = "<not" ]; then
-		echo "Failed: cycles not counted with --bpf-counters"
+	if [ "$bpf_instructions" = "<not" ]; then
+		echo "Failed: instructions not counted with --bpf-counters"
 		exit 1
 	fi
 }
@@ -41,29 +41,29 @@ check_counts()
 test_bpf_counters()
 {
 	printf "Testing --bpf-counters "
-	base_cycles=$(perf stat --no-big-num -e cycles -- $workload 2>&1 | awk '/cycles/ {print $1}')
-	bpf_cycles=$(perf stat --no-big-num --bpf-counters -e cycles -- $workload  2>&1 | awk '/cycles/ {print $1}')
-	check_counts $base_cycles $bpf_cycles
-	compare_number $base_cycles $bpf_cycles
+	base_instructions=$(perf stat --no-big-num -e instructions -- $workload 2>&1 | awk '/instructions/ {print $1}')
+	bpf_instructions=$(perf stat --no-big-num --bpf-counters -e instructions -- $workload  2>&1 | awk '/instructions/ {print $1}')
+	check_counts $base_instructions $bpf_instructions
+	compare_number $base_instructions $bpf_instructions
 	echo "[Success]"
 }
 
 test_bpf_modifier()
 {
 	printf "Testing bpf event modifier "
-	stat_output=$(perf stat --no-big-num -e cycles/name=base_cycles/,cycles/name=bpf_cycles/b -- $workload 2>&1)
-	base_cycles=$(echo "$stat_output"| awk '/base_cycles/ {print $1}')
-	bpf_cycles=$(echo "$stat_output"| awk '/bpf_cycles/ {print $1}')
-	check_counts $base_cycles $bpf_cycles
-	compare_number $base_cycles $bpf_cycles
+	stat_output=$(perf stat --no-big-num -e instructions/name=base_instructions/,instructions/name=bpf_instructions/b -- $workload 2>&1)
+	base_instructions=$(echo "$stat_output"| awk '/base_instructions/ {print $1}')
+	bpf_instructions=$(echo "$stat_output"| awk '/bpf_instructions/ {print $1}')
+	check_counts $base_instructions $bpf_instructions
+	compare_number $base_instructions $bpf_instructions
 	echo "[Success]"
 }
 
 # skip if --bpf-counters is not supported
-if ! perf stat -e cycles --bpf-counters true > /dev/null 2>&1; then
+if ! perf stat -e instructions --bpf-counters true > /dev/null 2>&1; then
 	if [ "$1" = "-v" ]; then
 		echo "Skipping: --bpf-counters not supported"
-		perf --no-pager stat -e cycles --bpf-counters true || true
+		perf --no-pager stat -e instructions --bpf-counters true || true
 	fi
 	exit 2
 fi
Signature · 1 line
2.43.0
maintainer Applied
#2 lore raw
In reply to [email protected]
Quoted 11 lines · On Tue, 25 Jun 2024 11:20:01 +0200, [email protected] wrote:

The test has been failing for some time when two separate runs of perf benchmarks are recorded for cycles events and their counts are compared, while once the recording was done with option --bpf-counters and once without it. It is expected that the count of the samples should be within a certain range, firstly the difference was set to be within 10%, which was then later raised to 20%. However, the test case keeps failing on certain architectures as recording the provided benchmark can produce completely different counts based on the current load of the system.

[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung