gc-stop-of-the-world

How does GC affect your Java application?

This question is straightforward but it is not easy to answer as there are so many factors to be looked at:

Nature of the application: is it a giant monolith enterprise application with a lot of data
and few concurent users. or is it a microservice with some APIs and serves many concurrent
requests from multiple clients?
GC types and other JVM tunings: the JVM does not have one single GC implementations.
In fact, it has many
and you can explicitly choose one via a JVM’s command line option.

Fortunately, there is a simple way to find out the answer: code it! This repo contains a small
Java application that mimics the memory consumption of a multi-threaded microservice and prints
out stats of memory usage as well as memory allocation latency:

It starts n threads (configured via -DnumThreads=n)
Each thread loops k times (configured via -DnumLoops=k). For each loop:
- Allocate a small block of memory (default 1Mb, configured via -DblockSize=sizeInBytes).
- De-reference the block (set the variable to null).
- Randomly sleep for up to t milliseconds (configured via -DsleepTime=t)

Memory usage and duration required to allocate each memory block are recorded and stats is printed out when all threads finish execution.

Below is the result of my own test with the following configurations:

Machine: GCP’s n1-standard-4
OS: CentOS Linux 7 (Core) - 3.10.0-957.27.2.el7.x86_64
Java: OpenJDK 64-Bit Server VM 18.9 (build 11.0.4+11-LTS, mixed mode, sharing)

Collectors to test: G1, Parallel, Serial, Shenandoah and Z (with -XX:+UnlockExperimentalVMOptions).
Each collector is executed 3 times with sleepTime of 10, 100 and 1000 milliseconds.

Test 1: numThreads=4, numLoops=1000, blockSize=1048576

All tests run with JVM options:

-server -XX:+ExitOnOutOfMemoryError -XX:+CrashOnOutOfMemoryError -Xms16m -Xmx1234m -XX:MinHeapFreeRatio=5 -XX:MaxHeapFreeRatio=10 -XX:-ShrinkHeapInStep

Latency Stats

Numbers are rounded to 3 decimal digits

Collector	Min (ms)	Max (ms)	Mean (ms)	p99 (ms)	p95 (ms)	p90 (ms)
G1, sleep 10	0.106	4.422	0.204	1.987	0.257	0.186
G1, sleep 100	0.128	4.117	0.296	2.175	1.701	0.243
G1, sleep 1000	0.131	9.125	0.883	2.376	1.974	1.898
Parallel, sleep 10	0.048	10.770	0.301	4.662	0.648	0.203
Parallel, sleep 100	0.065	11.465	0.336	4.314	1.267	0.812
Parallel, sleep 1000	0.128	6.353	0.476	1.729	1.555	1.320
Serial, sleep 10	0.061	6.736	0.226	0.894	0.481	0.412
Serial, sleep 100	0.123	3.286	0.256	0.695	0.576	0.545
Serial, sleep 1000	0.129	10.111	0.266	0.651	0.586	0.561
Shenandoah, sleep 10	0.098	13.331	0.497	3.641	2.258	1.441
Shenandoah, sleep 100	0.129	6.333	0.360	1.730	1.384	0.780
Shenandoah, sleep 1000	0.129	4.055	0.347	1.549	1.227	0.755
Z, sleep 10	0.119	26.713	0.393	1.724	0.699	0.620
Z, sleep 100	0.125	9.453	0.188	0.639	0.390	0.195
Z, sleep 1000	0.131	6.521	0.197	0.670	0.249	0.212

Memory Usage

Collector	Min (mb)	Max (mb)	Mean (mb)	p99 (mb)	p95 (mb)	p90 (mb)
G1, sleep 10	3.3	153.8	72.9	153.3	145.4	137.3
G1, sleep 100	3.2	31.3	16.0	29.4	29.3	27.4
G1, sleep 1000	3.2	7.5	4.4	5.5	5.5	5.5
Parallel, sleep 10	2.2	171.2	46.7	162.2	134.7	113.6
Parallel, sleep 100	2.3	172.7	31.8	153.7	115.2	88.4
Parallel, sleep 1000	3.3	6.5	5.4	6.5	6.5	6.5
Serial, sleep 10	2.2	14.4	9.0	14.4	13.3	12.4
Serial, sleep 100	2.2	14.5	9.6	14.5	13.5	13.5
Serial, sleep 1000	2.2	6.5	4.0	6.5	5.5	5.5
Shenandoah, sleep 10	4.8	706.8	291.5	694.3	655.3	605.2
Shenandoah, sleep 100	4.8	701.3	288.7	691.8	651.7	601.7
Shenandoah, sleep 1000	4.8	701.3	288.6	691.8	651.7	601.7
Z, sleep 10	40.0	1,002.0	442.3	968.0	876.0	814.0
Z, sleep 100	38.0	202.0	87.6	170.0	138.0	136.0
Z, sleep 1000	38.0	170.0	74.8	138.0	106.0	104.0

Test 2: numThreads=16, numLoops=1000, blockSize=1048576

All tests run with JVM options:

-server -XX:+ExitOnOutOfMemoryError -XX:+CrashOnOutOfMemoryError -Xms16m -Xmx1234m -XX:MinHeapFreeRatio=5 -XX:MaxHeapFreeRatio=10 -XX:-ShrinkHeapInStep

Latency Stats

Numbers are rounded to 3 decimal digits

Collector	Min (ms)	Max (ms)	Mean (ms)	p99 (ms)	p95 (ms)	p90 (ms)
G1, sleep 10	0.114	26.611	0.259	2.645	0.299	0.225
G1, sleep 100	0.121	11.647	0.209	1.975	0.230	0.179
G1, sleep 1000	0.127	12.080	0.360	2.095	1.841	1.145
Parallel, sleep 10	0.047	28.573	0.267	5.314	0.250	0.207
Parallel, sleep 100	0.064	14.955	0.175	0.571	0.174	0.159
Parallel, sleep 1000	0.076	11.991	0.262	1.890	0.980	0.213
Serial, sleep 10	0.041	33.446	1.044	10.133	6.464	2.800
Serial, sleep 100	0.097	14.490	0.269	0.968	0.576	0.525
Serial, sleep 1000	0.125	18.557	0.286	0.731	0.625	0.595
Shenandoah, sleep 10	0.121	74.189	0.813	19.124	1.907	0.243
Shenandoah, sleep 100	0.124	7.310	0.189	0.882	0.416	0.200
Shenandoah, sleep 1000	0.126	4.078	0.197	0.710	0.447	0.213
Z, sleep 10	0.113	89.410	0.335	0.929	0.620	0.233
Z, sleep 100	0.121	16.970	0.176	0.622	0.202	0.175
Z, sleep 1000	0.129	9.812	0.173	0.605	0.200	0.187

Memory Usage

Collector	Min (mb)	Max (mb)	Mean (mb)	p99 (mb)	p95 (mb)	p90 (mb)
G1, sleep 10	3.3	231.8	113.6	227.9	219.3	205.9
G1, sleep 100	3.3	76.0	38.3	74.0	71.4	67.6
G1, sleep 1000	3.3	31.6	10.6	18.0	18.0	17.5
Parallel, sleep 10	3.3	410.7	182.3	405.7	383.7	355.7
Parallel, sleep 100	2.3	410.7	201.1	406.7	389.5	367.7
Parallel, sleep 1000	2.3	406.6	116.6	370.1	318.6	281.4
Serial, sleep 10	3.3	18.4	11.1	14.6	14.4	13.6
Serial, sleep 100	3.3	18.4	11.1	14.6	14.4	13.6
Serial, sleep 1000	2.3	17.4	8.8	16.3	13.6	12.6
Shenandoah, sleep 10	5.2	720.2	352.8	704.2	674.2	636.2
Shenandoah, sleep 100	4.6	706.1	346.0	697.1	667.1	631.1
Shenandoah, sleep 1000	4.8	704.3	345.0	696.3	666.3	630.3
Z, sleep 10	38.0	1,098.0	268.9	680.0	490.0	424.0
Z, sleep 100	38.0	462.0	132.1	364.0	234.0	202.0
Z, sleep 1000	38.0	202.0	79.2	168.0	136.0	136.0

Test 3: numThreads=64, numLoops=1000, blockSize=1048576

All tests run with JVM options:

-server -XX:+ExitOnOutOfMemoryError -XX:+CrashOnOutOfMemoryError -Xms16m -Xmx1234m -XX:MinHeapFreeRatio=5 -XX:MaxHeapFreeRatio=10 -XX:-ShrinkHeapInStep

Latency Stats

Numbers are rounded to 3 decimal digits

Collector	Min (ms)	Max (ms)	Mean (ms)	p99 (ms)	p95 (ms)	p90 (ms)
G1, sleep 10	0.114	60.220	1.349	16.413	7.862	3.947
G1, sleep 100	0.117	29.938	0.219	2.093	0.255	0.205
G1, sleep 1000	0.123	42.617	0.257	2.053	1.066	0.197
Parallel, sleep 10	0.090	114.594	0.519	8.792	0.254	0.226
Parallel, sleep 100	0.047	127.121	0.244	0.355	0.195	0.170
Parallel, sleep 1000	0.066	53.084	0.191	0.280	0.179	0.169
Serial, sleep 10	0.051	279.767	26.657	150.036	106.703	85.245
Serial, sleep 100	0.043	135.802	0.775	13.512	1.118	0.627
Serial, sleep 1000	0.102	96.826	0.256	0.801	0.694	0.654
Shenandoah, sleep 10	0.113	1,217.492	3.818	98.935	0.273	0.235
Shenandoah, sleep 100	0.115	406.483	0.960	0.764	0.195	0.170
Shenandoah, sleep 1000	0.119	302.410	0.315	0.622	0.185	0.173
Z, sleep 10	0.108	209.871	0.813	11.563	4.138	0.580
Z, sleep 100	0.113	39.367	0.170	0.605	0.196	0.170
Z, sleep 1000	0.123	9.404	0.152	0.204	0.176	0.168

Memory Usage

Collector	Min (mb)	Max (mb)	Mean (mb)	p99 (mb)	p95 (mb)	p90 (mb)
G1, sleep 10	3.3	527.3	239.7	503.3	467.3	433.3
G1, sleep 100	3.3	131.3	67.3	128.8	123.3	117.3
G1, sleep 1000	3.3	101.4	26.5	47.3	45.3	43.3
Parallel, sleep 10	2.4	293.2	131.1	289.4	274.5	256.4
Parallel, sleep 100	2.3	411.7	205.7	407.1	390.7	369.8
Parallel, sleep 1000	2.3	411.7	207.6	407.1	390.8	370.4
Serial, sleep 10	0.8	91.8	23.2	67.5	31.8	29.0
Serial, sleep 100	0.4	53.5	11.5	48.4	14.7	13.8
Serial, sleep 1000	3.4	75.4	36.4	72.9	70.9	68.9
Shenandoah, sleep 10	4.8	730.3	362.7	705.3	672.3	637.3
Shenandoah, sleep 100	3.3	708.8	346.0	697.3	669.3	633.3
Shenandoah, sleep 1000	3.3	702.8	348.9	693.8	665.8	629.8
Z, sleep 10	46.0	1,228.0	663.5	1,196.0	1,166.0	1,106.0
Z, sleep 100	40.0	1,032.0	472.3	942.0	872.0	808.0
Z, sleep 1000	38.0	234.0	113.4	200.0	168.0	168.0

Summary

numThreads=4, numLoops=1000, blockSize=1048576

numThreads=16, numLoops=1000, blockSize=1048576

numThreads=64, numLoops=1000, blockSize=1048576