Here you will learn how to build and run the simplest example and measure the startup time on Linux using Debian packages
sudo apt install linux-tools-common linux-tools-generic
sudo sysctl kernel.perf_event_paranoid=1
scala-cli --power package --assembly example01.sc --force -o example01.jar
ls -l ./example01.jar
perf stat -r 100 ./example01.jar > /dev/nullExpected output:
Performance counter stats for './example01.jar' (100 runs):
202.75 msec task-clock # 2.286 CPUs utilized ( +- 0.27% )
2,191 context-switches # 10.806 K/sec ( +- 1.53% )
39 cpu-migrations # 192.356 /sec ( +- 2.49% )
24,775 page-faults # 122.195 K/sec ( +- 0.27% )
303,999,880 cpu_atom/cycles/ # 1.499 GHz ( +- 3.51% ) (71.71%)
1,005,678,639 cpu_core/cycles/ # 4.960 GHz ( +- 0.44% ) (77.24%)
431,508,692 cpu_atom/instructions/ # 1.42 insn per cycle ( +- 3.69% ) (71.71%)
1,684,651,998 cpu_core/instructions/ # 5.54 insn per cycle ( +- 0.47% ) (77.24%)
83,740,906 cpu_atom/branches/ # 413.027 M/sec ( +- 3.78% ) (71.71%)
335,230,949 cpu_core/branches/ # 1.653 G/sec ( +- 0.49% ) (77.24%)
2,017,405 cpu_atom/branch-misses/ # 2.41% of all branches ( +- 4.20% ) (71.71%)
9,450,892 cpu_core/branch-misses/ # 11.29% of all branches ( +- 0.47% ) (77.24%)
0.088700 +- 0.000135 seconds time elapsed ( +- 0.15% )
Build Scala JS output, print its size, and measure its start up time with node (tested with node.js 22)
curl https://raw.githubusercontent.com/creationix/nvm/master/install.sh | bash
source ~/.bashrc
nvm install 22
sudo sysctl kernel.perf_event_paranoid=1
scala-cli --power package --js --js-mode release example01.sc --force -o example01.js
ls -l ./example01.js
perf stat -r 100 node ./example01.js > /dev/nullExpected output:
Performance counter stats for 'node ./example01.js' (100 runs):
14.90 msec task-clock # 1.014 CPUs utilized ( +- 0.30% )
20 context-switches # 1.343 K/sec ( +- 1.60% )
2 cpu-migrations # 134.254 /sec ( +- 5.08% )
2,748 page-faults # 184.466 K/sec ( +- 0.00% )
45,371,808 cpu_atom/cycles/ # 3.046 GHz ( +- 2.61% ) (20.12%)
75,772,761 cpu_core/cycles/ # 5.086 GHz ( +- 0.32% ) (85.22%)
97,892,289 cpu_atom/instructions/ # 2.16 insn per cycle ( +- 3.38% ) (20.12%)
165,775,873 cpu_core/instructions/ # 3.65 insn per cycle ( +- 0.29% ) (85.22%)
16,573,542 cpu_atom/branches/ # 1.113 G/sec ( +- 3.35% ) (20.12%)
28,245,004 cpu_core/branches/ # 1.896 G/sec ( +- 0.27% ) (85.22%)
97,278 cpu_atom/branch-misses/ # 0.59% of all branches ( +- 11.08% ) (20.12%)
573,362 cpu_core/branch-misses/ # 3.46% of all branches ( +- 0.72% ) (85.22%)
0.0146864 +- 0.0000435 seconds time elapsed ( +- 0.30% )
Build Scala JS Wasm output, print its size, and measure its start up time with node (tested with node.js 22)
curl https://raw.githubusercontent.com/creationix/nvm/master/install.sh | bash
source ~/.bashrc
nvm install 22
sudo sysctl kernel.perf_event_paranoid=1
scala-cli --power package --js --js-mode release --js-emit-wasm --js-module-kind es --js-module-split-style fewestmodules example01.sc --force
ls -l ./example01.js/
perf stat -r 100 node --experimental-wasm-exnref ./example01.js/main.js > /dev/nullExpected output:
Performance counter stats for 'node --experimental-wasm-exnref ./example01.js/main.js' (100 runs):
23.36 msec task-clock # 1.036 CPUs utilized ( +- 0.23% )
61 context-switches # 2.611 K/sec ( +- 0.47% )
3 cpu-migrations # 128.413 /sec ( +- 4.53% )
3,156 page-faults # 135.090 K/sec ( +- 0.01% )
81,924,942 cpu_atom/cycles/ # 3.507 GHz ( +- 0.93% ) (10.81%)
115,425,581 cpu_core/cycles/ # 4.941 GHz ( +- 0.30% )
151,891,461 cpu_atom/instructions/ # 1.85 insn per cycle ( +- 1.65% ) (10.81%)
228,699,784 cpu_core/instructions/ # 2.79 insn per cycle ( +- 0.23% )
26,745,975 cpu_atom/branches/ # 1.145 G/sec ( +- 1.49% ) (10.81%)
39,592,483 cpu_core/branches/ # 1.695 G/sec ( +- 0.24% )
459,225 cpu_atom/branch-misses/ # 1.72% of all branches ( +- 1.76% ) (10.81%)
1,062,312 cpu_core/branch-misses/ # 3.97% of all branches ( +- 0.30% )
0.0225519 +- 0.0000530 seconds time elapsed ( +- 0.23% )
Build GraalVM native image, print its size, and measure its start up time (tested with Oracle GraalVM 24)
sudo apt install linux-tools-common linux-tools-generic gcc zlib1g-dev
sudo sysctl kernel.perf_event_paranoid=1
scala-cli --power package --graalvm-jvm-id graalvm-oracle:24 --native-image example01.sc --force -o example01_graalvm.bin -- --no-fallback -O3 -H:+UnlockExperimentalVMOptions -R:MaxHeapSize=16m -H:-GenLoopSafepoints -H:-ParseRuntimeOptions -H:-IncludeMethodData --initialize-at-build-time
ls -l ./example01_graalvm.bin
perf stat -r 100 ./example01_graalvm.bin > /dev/nullExpected output:
Performance counter stats for './example01_graalvm.bin' (100 runs):
1.08 msec task-clock # 0.911 CPUs utilized ( +- 0.73% )
1 context-switches # 927.547 /sec ( +- 4.79% )
0 cpu-migrations # 0.000 /sec
477 page-faults # 442.440 K/sec ( +- 0.01% )
4,315,458 cpu_atom/cycles/ # 4.003 GHz ( +- 1.85% ) (19.75%)
3,872,466 cpu_core/cycles/ # 3.592 GHz ( +- 5.11% )
10,829,542 cpu_atom/instructions/ # 2.51 insn per cycle ( +- 2.48% ) (19.75%)
5,351,563 cpu_core/instructions/ # 1.24 insn per cycle ( +- 5.21% )
1,856,118 cpu_atom/branches/ # 1.722 G/sec ( +- 2.33% ) (19.75%)
1,000,400 cpu_core/branches/ # 927.918 M/sec ( +- 5.18% )
7,651 cpu_atom/branch-misses/ # 0.41% of all branches ( +- 6.90% ) (19.75%)
11,792 cpu_core/branch-misses/ # 0.64% of all branches ( +- 5.28% )
0.00118361 +- 0.00000667 seconds time elapsed ( +- 0.56% )
Build Scala Native image, print its size, and measure its start up time (tested with Scala Native 0.5.7)
sudo apt install linux-tools-common linux-tools-generic clang libstdc++-12-dev libgc-dev
sudo sysctl kernel.perf_event_paranoid=1
scala-cli --power package --native-version 0.5.7 --native example01.sc --native-mode release-full --force -o example01_native.bin
ls -l ./example01_native.bin
perf stat -r 100 ./example01_native.bin > /dev/nullExpected output:
Performance counter stats for './example01_native.bin' (100 runs):
0.71 msec task-clock # 0.855 CPUs utilized ( +- 0.64% )
0 context-switches # 0.000 /sec
0 cpu-migrations # 0.000 /sec
731 page-faults # 1.024 M/sec ( +- 0.01% )
<not counted> cpu_atom/cycles/ ( +-100.00% ) (0.00%)
3,769,012 cpu_core/cycles/ # 5.280 GHz ( +- 1.17% )
<not counted> cpu_atom/instructions/ ( +-100.00% ) (0.00%)
5,907,572 cpu_core/instructions/ # 139.11 insn per cycle ( +- 1.02% )
<not counted> cpu_atom/branches/ ( +-100.01% ) (0.00%)
1,055,882 cpu_core/branches/ # 1.479 G/sec ( +- 1.02% )
<not counted> cpu_atom/branch-misses/ ( +-100.62% ) (0.00%)
9,104 cpu_core/branch-misses/ # 84.85% of all branches ( +- 1.35% )
0.00083490 +- 0.00000519 seconds time elapsed ( +- 0.62% )
An example of a command line application that reads the system input, parses and validates JSON according to the latest specification and in case of any error prints it to the system error output.
Scala.js build is missing because it doesn't support reading from the system input.
No GC options are used to speed up JSON validation for all supported builds.
Also, this script was tested with large (megabytes) and huge (gigabytes) inputs that could be downloaded and unpacked using following commands:
wget https://raw.githubusercontent.com/json-iterator/test-data/master/large-file.json
wget https://webtpa-public-access.s3.us-west-2.amazonaws.com/subfolder/2023_06_430_65B0_in_network_rates.json.gz
gunzip 2023_06_430_65B0_in_network_rates.json.gzSpeed of validation highly depends on performance of file reading on the particular disk system. Use the following command to measure reading speed on Linux:
sudo hdparm -Tt /dev/nvme0n1p4Here is an example of expected output:
/dev/nvme0n1p4:
Timing cached reads: 48588 MB in 2.00 seconds = 24334.91 MB/sec
Timing buffered disk reads: 4456 MB in 3.00 seconds = 1485.08 MB/sec
scala-cli --power package --assembly example02.sc --force -o example02.jar
ls -l ./example02.jar
time ./example02.jar -J-XX:+UnlockExperimentalVMOptions -J-XX:+UseEpsilonGC -J-Xms8m -J-Xmx8m -J-XX:+AlwaysPreTouch < 2023_06_430_65B0_in_network_rates.json 2> /dev/nullExpected output:
real 0m37.615s
user 0m35.187s
sys 0m2.662s
Build GraalVM native image, print its size, and measure its running time (tested with Oracle GraalVM 23)
scala-cli --power package --graalvm-jvm-id graalvm-oracle:23 --native-image example02.sc --force -o example02_graalvm.bin -- --no-fallback --gc=epsilon -O3 -H:+UnlockExperimentalVMOptions -R:MaxHeapSize=16m -H:-GenLoopSafepoints -H:-ParseRuntimeOptions -H:-IncludeMethodData --initialize-at-build-time
ls -l ./example02_graalvm.bin
time ./example02_graalvm.bin < 2023_06_430_65B0_in_network_rates.json 2> /dev/nullExpected output:
real 0m44.963s
user 0m42.471s
sys 0m2.486s
You can use profile guided optimization (PGO) to improve performance of Oracle GraalVM native image, for that you need:
- build an instrumented GraalVM native image with
--pgo-instrumentoption added:
scala-cli --power package --graalvm-jvm-id graalvm-oracle:23 --native-image example02.sc --force -o example02_graalvm_instrumented.bin -- --no-fallback --gc=epsilon -O3 --pgo-instrument -H:+UnlockExperimentalVMOptions -R:MaxHeapSize=16m -H:-GenLoopSafepoints -H:-ParseRuntimeOptions -H:-IncludeMethodData --initialize-at-build-time
ls -l ./example02_graalvm_instrumented.bin- run the instrumented image and collect the profile data:
time ./example02_graalvm_instrumented.bin < 2023_06_430_65B0_in_network_rates.json 2> /dev/null- build a PGO-optimized GraalVM native image with
--pgo=default.iprofoption added:
scala-cli --power package --graalvm-jvm-id graalvm-oracle:23 --native-image example02.sc --force -o example02_graalvm_optimized.bin -- --no-fallback --gc=epsilon -O3 --pgo=default.iprof -H:+UnlockExperimentalVMOptions -R:MaxHeapSize=16m -H:-GenLoopSafepoints -H:-ParseRuntimeOptions -H:-IncludeMethodData --initialize-at-build-time
ls -l ./example02_graalvm_optimized.bin- run the PGO-optimized image:
time ./example02_graalvm_optimized.bin < 2023_06_430_65B0_in_network_rates.json 2> /dev/nullExpected output:
real 0m43.674s
user 0m41.213s
sys 0m2.461s
scala-cli --power package --native-version 0.5.7 --native example02.sc --native-mode release-full --native-gc none --native-lto thin --native-multithreading=false --force -o example02_native.bin
ls -l ./example02_native.bin
time ./example02_native.bin < 2023_06_430_65B0_in_network_rates.json 2> /dev/nullExpected output:
real 0m49.232s
user 0m46.838s
sys 0m2.389s