GIGA-BYTE TECHNOLOGY CO., LTD. R282-Z90 319609 SPECjbb2015-MultiJVM max-jOPS
233669 SPECjbb2015-MultiJVM critical-jOPS
Tested by: GIGA-BYTE TECHNOLOGY CO., LTD. Test Sponsor: GIGA-BYTE TECHNOLOGY CO., LTD. Test location: New Taipei City 231, Taiwan Test date: July 17, 2019
SPEC license #: 9082 Hardware Availability: August-2019 Software Availability: May-2019 Publication: Wed Aug 7 19:56:20 EDT 2019
Benchmark Results Summary
 
Overall Throughput RT curve
Overall SUT (System Under Test) Description
VendorGIGA-BYTE TECHNOLOGY CO., LTD.
Vendor URLhttps://www.gigabyte.com/
System SourceSingle Supplier
System DesignationServer Rack
Total Systems1
All SUT Systems IdenticalYES
Total Nodes1
All Nodes IdenticalYES
Nodes Per System1
Total Chips2
Total Cores128
Total Threads256
Total Memory Amount (GB)1024
Total OS Images1
SW EnvironmentNon-virtual
 
Hardware hw_1
NameR282-Z90
VendorGIGA-BYTE TECHNOLOGY CO., LTD.
Vendor URLhttps://www.gigabyte.com/
AvailableAugust-2019
ModelR282-Z90
Form Factor2U
CPU NameAMD EPYC 7742
CPU Characteristics64 core, 2.25GHz, 256MB L3 Cache (Turbo Boost Technology up to 3.4 GHz)
Number of Systems1
Nodes Per System1
Chips Per System2
Cores Per System128
Cores Per Chip64
Threads Per System256
Threads Per Core2
VersionD13 07/13/2019
CPU Frequency (MHz)2250
Primary Cache32KB(I)+32KB(D) per core
Secondary Cache512KB (I+D) per core
Tertiary Cache256MB (I+D) on chip per chip
Other CacheNone
Disk1 x SAMSUNG SSD 833 DCT 1.92TB
File Systemext4
Memory Amount (GB)1024
# and size of DIMM(s)16 x 64 GB
Memory Details64 GB 4Rx4 PC4-3200AA-L
# and type of Network Interface Cards (NICs)2 x 1 Gbit LOMs
Power Supply Quantity and Rating (W)2 x DELTA DPS-1200AB-4 1200w
Other HardwareNone
Cabinet/Housing/EnclosureNone
Shared DescriptionNone
Shared CommentNone
Notes
  • NA: The test sponsor attests, as of date of publication, that CVE-2017-5754 (Meltdown) is mitigated in the system as tested and documented.
  • Yes: The test sponsor attests, as of date of publication, that CVE-2017-5753 (Spectre variant 1) is mitigated in the system as tested and documented.
  • Yes: The test sponsor attests, as of date of publication, that CVE-2017-5715 (Spectre variant 2) is mitigated in the system as tested and documented.
Other Hardware network_1
NameNone
VendorNone
Vendor URLNone
VersionNone
AvailableNone
BitnessNone
NotesNone
Operating System os_1
NameSUSE Linux Enterprise Server 15 SP1
VendorSUSE
Vendor URLhttp://suse.com/
Version4.12.14-195-default
AvailableMay-2019
Bitness64
NotesNone
Java Virtual Machine jvm_1
NameOracle Java SE 12.0.1
VendorOracle
Vendor URLhttp://oracle.com/
VersionJava HotSpot 64-bit Server VM, version 12.0.1
AvailableApr-2019
Bitness64
Notesnote
Other Software other_1
NameNone
VendorNone
Vendor URLNone
VersionNone
AvailableNone
BitnessNone
NotesNone
Hardware
OS Images os_Image_1(1)
Hardware Description hw_1
Number of Systems 1
SW Environment non-virtual
Tuning

BIOS Settings:

  • SMT Control set to Auto
  • IOMMU set to Enabled
  • NUMA nodes per socket set to NPS2
  • Determinism Control set to Manual
  • Determinism Slider set to Power
  • cTDP Control set to Manual
  • cTDP set to 240
  • Package Power Limit Control set to Manual
  • Package Power Limit set to 240
  • Enforce POR set to Accept
  • Overclock set to Enabled
  • Memory Clock Speed set to 1467MHz
  • L1 Stream HW Prefetcher set to Disable
  • L2 Stream HW Prefetcher set to Disable

Notes notes
OS Image os_Image_1
JVM Instances jvm_Ctr_1(1), jvm_Backend_1(8), jvm_TxInjector_1(8)
OS Image Description os_1
Tuning

  • cpupower -c all frequency-set -g performance
  • tuned-adm profile throughput-performance

  • echo 950000 > /proc/sys/kernel/sched_rt_runtime_us
  • echo 100000000 > /proc/sys/kernel/sched_latency_ns
  • echo 80000 > /proc/sys/kernel/sched_migration_cost_ns
  • echo 600000 > /proc/sys/kernel/sched_min_granularity_ns
  • echo 100000 > /proc/sys/kernel/sched_wakeup_granularity_ns

  • echo 10000 > /proc/sys/vm/dirty_expire_centisecs
  • echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
  • echo 40 > /proc/sys/vm/dirty_ratio
  • echo 10 > /proc/sys/vm/dirty_background_ratio
  • echo 10 > /proc/sys/vm/swappiness

  • echo 0 > /proc/sys/kernel/numa_balancing
  • echo always > /sys/kernel/mm/transparent_hugepage/enabled
  • echo always > /sys/kernel/mm/transparent_hugepage/defrag

  • Add cgroup_disable=memory,cpu,cpuacct,blkio,hugetlb,pids,cpuset,perf_event,freezer,devices,net_cls,net_prio to GRUB_CMDLINE_LINUX_DEFAULT
  • ulimit -n 1024000
  • UserTasksMax=970000
  • DefaultTasksMax=970000

Notes None
JVM Instance jvm_Ctr_1
Parts of Benchmark Controller
JVM Instance Description jvm_1
Command Line

-Xms2g -Xmx2g -Xmn1536m -XX:+UseParallelOldGC -XX:ParallelGCThreads=2 -XX:CICompilerCount=4

Tuning None
Notes

Used numactl to interleave memory on all CPUs

  • numactl --interleave=all

JVM Instance jvm_Backend_1
Parts of Benchmark Backend
JVM Instance Description jvm_1
Command Line

-Xms118g -Xmx118g -Xmn116g -server -XX:MetaspaceSize=256m -XX:AllocatePrefetchInstr=2 -XX:LargePageSizeInBytes=2m -XX:-UsePerfData -XX:-UseAdaptiveSizePolicy -XX:+AlwaysPreTouch -XX:-UseBiasedLocking -XX:+UseLargePages -XX:+UseParallelOldGC -XX:SurvivorRatio=50 -XX:TargetSurvivorRatio=95 -XX:ParallelGCThreads=32 -XX:MaxTenuringThreshold=15 -XX:InitialCodeCacheSize=25m -XX:InlineSmallCode=10k -XX:MaxGCPauseMillis=200 -XX:+UseCompressedOops -XX:ObjectAlignmentInBytes=32 -XX:+BindGCTaskThreadsToCPUs -XX:+UseTransparentHugePages

Tuning None
Notes

Used numactl to affinitize two Backend JVM (2 Groups) to one NUMA node

  • Group1: numactl --cpunodebind=0 --localalloc
  • Group2: numactl --cpunodebind=0 --localalloc
  • Group3: numactl --cpunodebind=1 --localalloc
  • Group4: numactl --cpunodebind=1 --localalloc
  • Group5: numactl --cpunodebind=2 --localalloc
  • Group6: numactl --cpunodebind=2 --localalloc
  • Group7: numactl --cpunodebind=3 --localalloc
  • Group8: numactl --cpunodebind=3 --localalloc

JVM Instance jvm_TxInjector_1
Parts of Benchmark TxInjector
JVM Instance Description jvm_1
Command Line

-Xms2g -Xmx2g -Xmn1536m -XX:+UseParallelOldGC -XX:ParallelGCThreads=2 -XX:CICompilerCount=4

Tuning None
Notes

Used numactl to affinitize two Transaction Injector JVM (2 Groups) to one NUMA node

  • Group1: numactl --cpunodebind=0 --localalloc
  • Group2: numactl --cpunodebind=0 --localalloc
  • Group3: numactl --cpunodebind=1 --localalloc
  • Group4: numactl --cpunodebind=1 --localalloc
  • Group5: numactl --cpunodebind=2 --localalloc
  • Group6: numactl --cpunodebind=2 --localalloc
  • Group7: numactl --cpunodebind=3 --localalloc
  • Group8: numactl --cpunodebind=3 --localalloc

max-jOPS = jOPS passed before the First Failure
Pass/Fail Pass Pass Fail Fail Fail
jOPS 315663 319609 323555 327501 331446
critical-jOPS = Geomean ( jOPS @ 10000; 25000; 50000; 75000; 100000; SLAs )
Response time percentile is 99-th
SLA (us) 10000 25000 50000 75000 100000 Geomean
jOPS 175076 225473 248831 262395 270287 233669
  Percentile
  10-th 50-th 90-th 95-th 99-th 100-th
500us 3946 / 7892 - / 3946 - / 3946 - / 3946 - / 3946 - / 3946
1000us 126265 / 130211 15783 / 19729 3946 / 7892 3946 / 7892 - / 3946 - / 3946
5000us 291988 / 295934 276205 / 280151 220964 / 224910 189398 / 193344 114428 / 78916 - / 3946
10000us 295934 / 299880 288043 / 291988 268314 / 272260 260422 / 264368 213073 / 110482 31566 / 3946
25000us 295934 / 299880 291988 / 295934 280151 / 284097 272260 / 276205 236747 / 157832 31566 / 11837
50000us 299880 / 303826 291988 / 295934 284097 / 288043 276205 / 280151 252531 / 193344 43404 / 11837
75000us 299880 / 303826 291988 / 295934 284097 / 288043 280151 / 284097 260422 / 264368 43404 / 11837
100000us 303826 / 307772 291988 / 295934 284097 / 288043 284097 / 288043 268314 / 272260 43404 / 11837
200000us 303826 / 307772 295934 / 299880 291988 / 295934 288043 / 291988 280151 / 284097 240693 / 74970
500000us 319609 / - 303826 / 307772 295934 / 299880 295934 / 299880 291988 / 295934 288043 / 291988
1000000us 319609 / - 315663 / 319609 307772 / 311717 307772 / 311717 303826 / 307772 291988 / 295934
Probes jOPS / Total jOPS
Request Mix Accuracy
Note
(Actual % in the Mix - Expected % in the Mix) must be within:
'Main Tx' limit of +/-5.0% for the requests whose expected % in the mix is >= 10.0%
'Minor Tx' limit of +/-1.0% for the requests whose expected % in the mix is < 10.0%
There were no non-critical failures in Response Time curve building
Delay between status pings
IR/PR Accuracy
This section lists properties only set by user
Property Name Default Controller
specjbb.controller.handshake.period 5000 20000
specjbb.controller.handshake.timeout 600000 800000
specjbb.forkjoin.workers 256 {Tier1=64, Tier2=1, Tier3=12}
specjbb.group.count 1 8
specjbb.heartbeat.period 10000 2000
specjbb.heartbeat.threshold 100000 900000
specjbb.txi.pergroup.count 1 1
View table in csv format
 
Level: COMPLIANCE
Check Agent Result
Check properties on compliance All PASSED
 
Level: CORRECTNESS
Check Agent Result
Compare SM and HQ Inventory All PASSED
High-bound (max attempted) is 394579 IR
High-bound (settled) is 339220 IR