This repository contains main classes with the following examples:
- A command line utility
- Simple stress test to measure the performance of any given configuration
- A multi-threaded test that monitors memory consumption for any given configuration (Linux only)
All examples use the default Jargon2 backend implementation with generic binaries.
Class com.kosprov.jargon2.examples.CommandLineUtility mimics the command line utility implemented in Argon2 reference implementation.
This class is the Main-Class of the (fat) jar produced, so it can be executed as:
mvn clean package;
echo -n "password" | java -jar target/jargon2-examples-1.1.1.jar somesalt -t 2 -m 16 -p 4 -l 24The output is:
Type: Argon2i
Iterations: 2
Memory: 65536 KiB
Parallelism: 4
Hash: 45d7ac72e76f242b20b77b9bf9bf9d5915894e669a24e6c6
Encoded: $argon2i$v=19$m=65536,t=2,p=4$c29tZXNhbHQ$RdescudvJCsgt3ub+b+dWRWJTmaaJObG
0.134 seconds
Verification ok
Class com.kosprov.jargon2.examples.StressTest implements a microbenchmark of Jargon2.
mvn clean package;
java -server -Xms512m -Xmx512m -cp target/jargon2-examples-1.1.1.jar com.kosprov.jargon2.examples.StressTest \
--iterations 100 \
--outputType encoded \
--saltLength 16 \
--passwordLength 64 \
--secretLength 32 \
--adLength 64 \
--hashLength 16 \
--type id \
--version 13 \
--memoryCost $(bc <<< "32 * 1024") \
--timeCost 3 \
--parallelism 1 \
;The output is:
Output type: encoded
Salt length: 16 bytes
Password length: 64 bytes
Secret length: 32 bytes
AD length: 64 bytes
Hash length: 16 bytes
Type: Argon2id
Version: v13
Memory cost: 32768 KB
Time cost: 3 passes
Parallelism: 1 lanes/threads
--------------------------------------------------
Warming up...
Running stress test...
Iterations: 100, Output type: encoded, AD length: 64, Salt length: 16, Password length: 64
Hasher{backend=com.kosprov.jargon2.backend.NativeRiJargon2Backend, options=none, type=ARGON2id, version=V13, timeCost=3, memoryCost=32768, lanes=1, threads=1, hashLength=16, saltLength=16}
Verifier{backend=com.kosprov.jargon2.backend.NativeRiJargon2Backend, options=none, type=ARGON2id, version=V13, timeCost=3, memoryCost=32768, lanes=1, threads=1}
[ avg min max (95%) ]
Hash : [ 66.47ms, 64884716ns, 68156945ns]
Verify: [ 66.34ms, 65167428ns, 67423742ns]
Total : 12617ms
Folder scripts/stress-tests contains a few shell scripts for different configurations.
Class com.kosprov.jargon2.examples.MultiThreadedHashVerifyLoop implements a multi-threaded hash/verify loop while a separate thread measures heap and process memory consumption.
The idea is to be able to run for a significant number of hash/verify operations with different memory requirements (both the memory cost and the data passed) in order to spot any memory leaks or stability issues.
Memory consumption at the process level is measured with
psso it's just an indication, not an accurate measurement.
For example, 4 Java threads executing a hash/verify loop for 2 hours with moderate settings for values and Argon2 configuration would look like:
mvn clean package;
java -Xms128m -Xmx128m -cp target/jargon2-examples-1.1.1.jar com.kosprov.jargon2.examples.MultiThreadedHashVerifyLoop \
--runtime 7200 \
--collectStats 60 \
--javaThreads 4 \
--saltLength 16 \
--passwordLength 12 \
--secretLength 0 \
--adLength 0 \
--hashLength 32 \
--type id \
--version 13 \
--memoryCost 32768 \
--timeCost 1 \
--parallelism 1The output is:
--------------------------------------------------
Configuration
--------------------------------------------------
Backend: automatic
Runtime: 7200 seconds
Collect stats: every 60 seconds
Java threads: 4
Salt length: 16 bytes
Password length: 12 bytes
Secret length: 0 bytes
AD length: 0 bytes
Hash length: 32 bytes
Type: Argon2id
Version: v13
Memory cost: 32768 KB
Time cost: 1 passes
Parallelism: 1 lanes/threads
--------------------------------------------------
[10:49:21] Pid: 21958. Estimated completion before 12:50:21
T,C,%CPU,RSS,S0C,S1C,S0U,S1U,EC,EU,OC,OU,MC,MU,CCSC,CCSU,YGC,YGCT,FGC,FGCT,GCT
0,0,0.0,32860,5120.0,5120.0,0.0,0.0,33280.0,3994.8,87552.0,0.0,4480.0,773.8,384.0,75.8,0,0.000,0,0.000,0.000
60284,2822,358,147788,5120.0,5120.0,0.0,0.0,33280.0,31286.6,87552.0,0.0,4480.0,773.8,384.0,75.8,0,0.000,0,0.000,0.000
120513,5730,361,171928,5120.0,5120.0,0.0,2559.1,33280.0,20782.9,87552.0,8.0,7040.0,6758.5,896.0,745.7,1,0.018,0,0.000,0.018
180723,8592,362,122136,5120.0,5120.0,2464.0,0.0,33280.0,8997.7,87552.0,16.0,7040.0,6768.4,896.0,745.7,2,0.043,0,0.000,0.043
... 2 hours later ...
7166095,331780,361,182484,2048.0,2048.0,0.0,1664.0,39424.0,32582.2,87552.0,2200.1,7296.0,6869.6,896.0,746.7,63,0.779,0,0.000,0.779
7226401,333328,360,107384,2560.0,2560.0,0.0,0.0,38400.0,28.2,87552.0,1039.6,7296.0,6864.7,896.0,744.0,67,0.803,3,0.043,0.846
[12:49:47] Executed 333328 hash/verify in 7200s.
A plot of some of these metrics is:
Folder scripts/long-running contains a few shell scripts for different configurations that could potentially expose a stability issue. Also, there are spreadsheets where you can paste the CSV data and reproduce the plot.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent