Experiments exploring the potential benefits of cluster aware scheduling/worker affinity for virtual threads. Developed against 24-loom+7-60.
See https://mail.openjdk.org/pipermail/loom-dev/2024-September/007161.html for background.
Benchmarks run on an AWS EC2 m7a.8xlarge instance:
# JMH version: 1.37
# VM version: JDK 24-loom, OpenJDK 64-Bit Server VM, 24-loom+7-60
# VM invoker: /usr/lib/jvm/openjdk-24-loom/bin/java
# VM options: -Dfile.encoding=UTF-8 -Duser.country=US -Duser.language=en -Duser.variant
External submission throughput:
Benchmark (placement) Mode Cnt Score Error Units
ExternalSubmission.submit FJP thrpt 5 64702.102 ± 1615.238 ops/s
ExternalSubmission.submit CHOOSE_TWO thrpt 5 79286.111 ± 1336.440 ops/s
ExternalSubmission.submit ROUND_ROBIN thrpt 5 86255.203 ± 729.912 ops/s
ExternalSubmission.submit CURRENT thrpt 5 102784.334 ± 13334.137 ops/s
Benchmark (multiplier) (scheduler) Mode Cnt Score Error Units
VirtualThreadsSchedulerCacheStress.submit 1 CHOOSE_TWO thrpt 5 4873.550 ± 290.117 ops/min (~1750% CPU)
VirtualThreadsSchedulerCacheStress.submit 1 DEFAULT thrpt 5 3782.106 ± 69.471 ops/min (~2650% CPU)
VirtualThreadsSchedulerCacheStress.submit 10 CHOOSE_TWO thrpt 5 453.875 ± 80.840 ops/min
VirtualThreadsSchedulerCacheStress.submit 10 DEFAULT thrpt 5 324.959 ± 25.635 ops/min
VirtualThreadsSchedulerCacheStress.submit 100 CHOOSE_TWO thrpt 5 44.731 ± 23.493 ops/min
VirtualThreadsSchedulerCacheStress.submit 100 DEFAULT thrpt 5 37.416 ± 5.211 ops/min
processor : 0
vendor_id : AuthenticAMD
cpu family : 25
model : 17
model name : AMD EPYC 9R14
stepping : 1
microcode : 0xa101148
cpu MHz : 3698.291
cache size : 1024 KB
physical id : 0
siblings : 32
core id : 0
cpu cores : 32
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 16
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext perfctr_core invpcid_single ssbd perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx512_bf16 clzero xsaveerptr rdpru wbnoinvd arat avx512vbmi pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid flush_l1d
bugs : sysret_ss_attrs spectre_v1 spectre_v2 spec_store_bypass srso
bogomips : 5200.00
TLB size : 3584 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 48 bits physical, 48 bits virtual
power management: