1 个不稳定版本
| 0.1.0 | 2023年10月6日 |
|---|
#76 in #agent
15KB
resctrl 代理
这是一个用于将 JVM 放入 COS (服务类别) 的包装共享库，使用 Linux resctrl 接口。共享库由两个环境变量控制，即 RESCTRL_COS 和 RESCTRL_SCHEMATA;
库将在 /sys/fs/resctrl/$RESCTRL_COS 下运行。如果当前进程是 java (即，argv[0] 以 java 结尾)，则 PID 将写入 /sys/fs/resctrl/$RESCTRL_COS/tasks，所有子线程将自动添加到 COS 中。否则，代理将不执行任何操作。
如果您设置了 RESCTRL_SCHEMATA，则内容将写入 /sys/fs/resctrl/$RESCTRL_COS/schemata。否则，将保留当前方案。
当进程退出时，如果代理处于活动状态（即，与 JVM 运行），则将清除 /sys/fs/resctrl/$RESCTRL_COS/tasks 中的内容。
您需要确保当前用户是 /sys/fs/resctrl/$RESCTRL_COS/tasks 和 /sys/fs/resctrl/$RESCTRL_COS/schemata 的所有者，以便可以写入这些文件。
示例
设置
sudo mount -t resctrl resctrl /sys/fs/resctrl
cd /sys/fs/resctrl
sudo mkdir dacapo_cos
cd dacapo_cos
sudo chown zixianc:zixianc tasks
sudo chown zixianc:zixianc schemata
构建
cargo build --release
运行程序
$ RESCTRL_COS="dacapo_cos" RESCTRL_SCHEMATA="L3:0=0000;1=0000" LD_PRELOAD=$PWD/target/release/libresctrl.so perf stat /usr/lib/jvm/temurin-
17-jdk-amd64/bin/java -jar /usr/share/benchmarks/dacapo/dacapo-23.9-RC3-chopin.jar h2 -n 3
Operating on COS dacapo_cos
Setting schemata to L3:0=0000;1=0000
Setting tasks to 1258882
Hello 1258882, schemata MB:0=2048;1=2048
L3:0=0000;1=0000
, tasks 1258882
Using scaled threading model. 32 processors detected, 32 threads used to drive the workload, in a possible range of [1,1024]
Version: h2 2.2.220 derby 10.15.2.0 (use -p to print nominal benchmark stats)
Preparing the H2 database...
TPCC configuration: scale: 16, terminals: 32, transactions: 100000
===== DaCapo 23.9-RC3-chopin h2 starting warmup 1 =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 completed warmup 1 in 6878 msec =====
===== DaCapo simple tail latency: 50% 602 usec, 90% 5951 usec, 99% 18979 usec, 99.9% 35013 usec, 99.99% 55585 usec, max 89408 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 792915 usec, 90% 1039091 usec, 99% 1049864 usec, 99.9% 1058446 usec, 99.99% 1067496 usec, max 1085802 usec, measured over 100000 events =====
===== DaCapo 23.9-RC3-chopin h2 starting warmup 2 =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 completed warmup 2 in 4718 msec =====
===== DaCapo simple tail latency: 50% 417 usec, 90% 3939 usec, 99% 13867 usec, 99.9% 25405 usec, 99.99% 35936 usec, max 48642 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 106652 usec, 90% 207804 usec, 99% 238194 usec, 99.9% 244516 usec, 99.99% 254887 usec, max 266923 usec, measured over 100000 events =====
===== DaCapo 23.9-RC3-chopin h2 starting =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 PASSED in 4509 msec =====
===== DaCapo simple tail latency: 50% 382 usec, 90% 3622 usec, 99% 13813 usec, 99.9% 24198 usec, 99.99% 34614 usec, max 47028 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 665 usec, 90% 14654 usec, 99% 25503 usec, 99.9% 34197 usec, 99.99% 44125 usec, max 59935 usec, measured over 100000 events =====
Operating on COS dacapo_cos
Clearing tasks
Performance counter stats for '/usr/lib/jvm/temurin-17-jdk-amd64/bin/java -jar /usr/share/benchmarks/dacapo/dacapo-23.9-RC3-chopin.jar h2 -n 3':
379,003.50 msec task-clock # 13.186 CPUs utilized
150,277 context-switches # 396.506 /sec
23,173 cpu-migrations # 61.142 /sec
2,180,511 page-faults # 5.753 K/sec
1,276,062,771,368 cycles # 3.367 GHz (83.29%)
4,364,759,840 stalled-cycles-frontend # 0.34% frontend cycles idle (83.33%)
82,311,638,243 stalled-cycles-backend # 6.45% backend cycles idle (83.39%)
847,207,831,918 instructions # 0.66 insn per cycle
# 0.10 stalled cycles per insn (83.40%)
162,966,950,783 branches # 429.988 M/sec (83.34%)
2,309,449,633 branch-misses # 1.42% of all branches (83.29%)
28.742477886 seconds time elapsed
368.652118000 seconds user
10.141901000 seconds sys
$ RESCTRL_COS="dacapo_cos" RESCTRL_SCHEMATA="L3:0=ffff;1=ffff" LD_PRELOAD=$PWD/target/release/libresctrl.so perf stat /usr/lib/jvm/temurin-
17-jdk-amd64/bin/java -jar /usr/share/benchmarks/dacapo/dacapo-23.9-RC3-chopin.jar h2 -n 3
Operating on COS dacapo_cos
Setting schemata to L3:0=ffff;1=ffff
Setting tasks to 1259428
Hello 1259428, schemata MB:0=2048;1=2048
L3:0=ffff;1=ffff
, tasks 1259428
Using scaled threading model. 32 processors detected, 32 threads used to drive the workload, in a possible range of [1,1024]
Version: h2 2.2.220 derby 10.15.2.0 (use -p to print nominal benchmark stats)
Preparing the H2 database...
TPCC configuration: scale: 16, terminals: 32, transactions: 100000
===== DaCapo 23.9-RC3-chopin h2 starting warmup 1 =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 completed warmup 1 in 3585 msec =====
===== DaCapo simple tail latency: 50% 322 usec, 90% 3050 usec, 99% 10479 usec, 99.9% 21308 usec, 99.99% 30042 usec, max 54112 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 367590 usec, 90% 571691 usec, 99% 612931 usec, 99.9% 617580 usec, 99.99% 622906 usec, max 627753 usec, measured over 100000 events =====
===== DaCapo 23.9-RC3-chopin h2 starting warmup 2 =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 completed warmup 2 in 2788 msec =====
===== DaCapo simple tail latency: 50% 251 usec, 90% 1939 usec, 99% 9236 usec, 99.9% 19570 usec, 99.99% 27703 usec, max 38695 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 9810 usec, 90% 35904 usec, 99% 44756 usec, 99.9% 50006 usec, 99.99% 56204 usec, max 64875 usec, measured over 100000 events =====
===== DaCapo 23.9-RC3-chopin h2 starting =====
Starting 100000 requests...
100%
Completed requests
Stock level ............. 3973 ( 4.0%)
Order status by name .... 2361 ( 2.4%)
Order status by ID ...... 1609 ( 1.6%)
Payment by name ......... 25847 (25.8%)
Payment by ID ........... 17220 (17.2%)
Delivery schedule ....... 4029 ( 4.0%)
New order ............... 44509 (44.5%)
New order rollback ...... 452 ( 0.5%)
===== DaCapo 23.9-RC3-chopin h2 PASSED in 2788 msec =====
===== DaCapo simple tail latency: 50% 243 usec, 90% 1921 usec, 99% 9175 usec, 99.9% 19241 usec, 99.99% 29108 usec, max 34147 usec, measured over 100000 events =====
===== DaCapo metered tail latency: 50% 260 usec, 90% 2070 usec, 99% 9198 usec, 99.9% 19260 usec, 99.99% 29108 usec, max 34147 usec, measured over 100000 events =====
Operating on COS dacapo_cos
Clearing tasks
Performance counter stats for '/usr/lib/jvm/temurin-17-jdk-amd64/bin/java -jar /usr/share/benchmarks/dacapo/dacapo-23.9-RC3-chopin.jar h2 -n 3':
210,033.51 msec task-clock # 10.480 CPUs utilized
129,766 context-switches # 617.835 /sec
17,137 cpu-migrations # 81.592 /sec
2,693,834 page-faults # 12.826 K/sec
708,207,359,243 cycles # 3.372 GHz (83.30%)
3,712,326,657 stalled-cycles-frontend # 0.52% frontend cycles idle (83.37%)
42,684,433,254 stalled-cycles-backend # 6.03% backend cycles idle (83.31%)
834,079,372,035 instructions # 1.18 insn per cycle
# 0.05 stalled cycles per insn (83.50%)
161,719,077,903 branches # 769.968 M/sec (83.27%)
2,160,291,673 branch-misses # 1.34% of all branches (83.30%)
20.041644681 seconds time elapsed
203.553157000 seconds user
6.357287000 seconds sys
依赖项
~290–790KB
~19K SLoC