A LinuxONE III Story The IBM z15 Chip is Awesome - VM ...
-
Upload
khangminh22 -
Category
Documents
-
view
0 -
download
0
Transcript of A LinuxONE III Story The IBM z15 Chip is Awesome - VM ...
Topics
A LinuxONE III StoryMonte Bauman
LinuxONE Technical Support Specialist
IBM Columbus
June 2021
The IBM z15 Chip is Awesome
IBM LinuxONE
© 2021 IBM Corporation4
LinuxONE
AbstractThe z15 chips is awesome. What follows are 9.1 Billion reasons why.
Moore’s Law used to be awesome…. and sort-of still is.
§ “Atoms ain’t getting smaller” … and “the speed of light ain’t getting faster”• And hence Moore’s Law is “underperforming”
§ But the demand for ever bigger and faster computing systems remains• Consumers know what they want … they just don’t know what they are asking for
§ Computer scientists and engineers must meet design challenges in new ways• Bigger problems requires bigger thinking
§ The IBM Z chip and the platforms it powers ( IBM Z and LinuxONE ) have been “stepping up” to the challenge and this presentation will illustrate strides made in hardware and software to make the platform excel
The numbers, they do tell a tale. This talk will examine the evolution of CMOS microprocessor technology on the mainframe platform. We will examine the intriguing relationship between MIPS and MHz, and between transistors, chips, and cores. We will observe the past and current effects of “Moore’s Law”, and we will look at equations (yes, equations!) suggesting how to combat computer science physics. Microprocessor engineers have quite a challenge ahead, what will they think of next?
© 2021 IBM Corporation5
LinuxONE
Monte Bauman … [email protected]
IBM Technical Support Specialist - helping clients build risk mitigated operationally excellent economic server solutions.
• Monte Bauman helps client's design and deploy risk mitigated operationally excellent super-hyperconverged server solutions using IBM LinuxONE technologies provided by IBM and by the LinuxONE partner ecosystem.
• Monte Bauman, Certified IT Specialist and z Systems Technical Support Professional, began a career in IT in 1983 hiring into the IBM Glendale Lab where he assisted a development team to build and test mid-range mainframe servers (4300s, 9370s, 9221s). Monte was given a Division Award for work on STX/370, a diagnostic operating system. Monte Bauman became a Large Systems Systems Engineer (SE) in January of 1991 in Columbus Ohio working with enterprise clients specializing in 9021 systems support, enterprise printing support (3800/3900), and MVS support. In the mid 90's Monte Bauman became a mainframe seller for a time, before returning in late 90's to the role of a regionally designated specialist large systems SE. Monte Bauman became a "new workloads" specialist at the turn of the century, technically supporting customers as they embarked upon emerging technologies on the mainframe, such as Java and Business Intelligence and Linux. The past several years Monte Bauman has been focused on LinuxONE and Fit for Purpose analysis, helping customers understand the fit of z Systems Linux solutions as well as the processes appropriate to repeatedly and effectively map software to hardware meeting IT optimization criteria for requirements and cost.
© 2021 IBM Corporation9
LinuxONE
z15Core
De-code De-Order Re-Order
Instruction Pointer
PipelineDispatch
Registers
Out of Order
Instruction Execution
© 2021 IBM Corporation10
LinuxONE
z15Core
De-code De-Order Re-Order
Instruction Pointer
PipelineDispatch
27Execution
Units
Registers
27 Execution
Units
Super-Scalar
© 2021 IBM Corporation11
LinuxONE
z15Core
De-code De-Order Re-Order
Instruction Pointer
PipelineDispatch
27Execution
Units
Registers
L1256KB
Big & Smart Cache
© 2021 IBM Corporation12
LinuxONE
z15Core
De-code De-Order Re-Order
Instruction Pointer
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
Big & Smart Cache
Big: Instruction
CacheBig: Data
Cache
Smart:Sophisticated
“cache associativity”
Smart:Software can provide tips to
the caches
© 2021 IBM Corporation13
LinuxONE
z15Core
De-code De-Order Re-Order
Instruction Pointer
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
TLB
Multi-Threaded
TLBAddress
Translation Lookaside
Buffer
Caching Addresses
© 2021 IBM Corporation14
LinuxONE
z15Core
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Load/Store Unit
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
TLB
Super-Scalar
Instructions can be categorized
by function performed, and hardware can be optimized
© 2021 IBM Corporation15
LinuxONE
z15Core
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
TLBSuper-Scalar
Optimize hardware for
the most common
instructions
© 2021 IBM Corporation16
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
TLB
Super-Scalar
Branches kill pipelines …
Unless!
© 2021 IBM Corporation17
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
27Execution
Units
Registers
L1256KB
L28MB
TLB
Branch Prediction
Cache Branch targets
Smartly influence caching
decisions
© 2021 IBM Corporation18
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
27Execution
Units
Modulo Arithmetic Modulo Arithmetic
Registers
L1256KB
L28MB
TLB
Workload Accelerators
Encryption
Compression
Put tough software
problems in hardware
Sorting
© 2021 IBM Corporation19
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
SIMDUnit2X8
1
8
2
16
Modulo Arithmetic Modulo Arithmetic
SIMD
Registers
L1256KB
L28MB
TLB
Single Instruction Multiple Data(aka Vector Processing)
© 2021 IBM Corporation20
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
Instruction Pointer
SIMDUnit2X8
1
8
2
16
Modulo Arithmetic Modulo Arithmetic
Registers Registers
L1256KB
L28MB
TLB
Simultaneous Multi-Threading with 1 Thread
SMT1
© 2021 IBM Corporation21
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
Instruction Pointer
SIMDUnit2X8
1
8
2
16
Modulo Arithmetic Modulo Arithmetic
Registers Registers
L1256KB
L28MB
TLB
Simultaneous Multi-Threading with 2 Threads
SMT2
© 2021 IBM Corporation22
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
Instruction Pointer
SIMDUnit2X8
BFUDFUDFxFPdVXxVXsVXpVXm
BFUDFUDFxFPdVXxVXsVXpVXm
Modulo Arithmetic Modulo Arithmetic
Registers Registers
IBM z Architecture
alive and well … hundreds of
new instructions
L1256KB
L28MB
New Instructions:20% more
throughput for general Java
workloads just from new instructions in
the z15 chip architecture TLB
© 2021 IBM Corporation23
LinuxONE
z15Core
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Retry
Instruction Pointer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
PipelineDispatch
Instruction Pointer
SIMDUnit2X8
BFUDFUDFxFPdVXxVXsVXpVXm
BFUDFUDFxFPdVXxVXsVXpVXm
Modulo Arithmetic Modulo Arithmetic
Error detection
and instruction
retry
Dynamic core
sparing
Phone Home
100 years
++ MTBF
Registers Registers
L1256KB
L28MB
First failure data
capture
TLB
© 2021 IBM Corporation24
LinuxONE5.2GHz
Out of
Order
Super-Scalar
27 Execution
Units
Big & Smart Cache
Branch Prediction
SIMD
Workload Accelerators
SMT2
Multi-Threaded
TLB
2,250***MIPS*
z15Core
General Purpose Registers
Branch Unit
Load/Store UnitDe-code De-Order Re-Order
Retry
General Purpose Registers
Floating Point Registers
Control Registers
Processor Status Word
Instruction Pointer
L1 InstructionCache 128KB
L1 DataCache 128KB
L2 Instruction Cache 4MB Translation
LookasideBuffer
Fixed Point UnitFixed Point Unit Branch
HistoryTable
Compression, Encryption (CPACF) & Sort
L2 DataCache 4MB
Branch Unit
Load/Store Unit
Fixed Point UnitFixed Point Unit
Pipeline
DAT
DAT DAT
DAT
Dispatch
General Purpose RegistersGeneral Purpose Registers
Floating Point Registers
Control Registers
Processor Status Word
Instruction Pointer
SIMDUnit2X8
BFUDFUDFxFPdVXxVXsVXpVXm
BFUDFUDFxFPdVXxVXsVXpVXm
Modulo Arithmetic Modulo Arithmetic
50 years ++
MTBF
© 2021 IBM Corporation27
LinuxONE
z15ChipCore
12 Cores
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
© 2021 IBM Corporation28
LinuxONE
z15Chip
L3128MB
L3128MB
L1/L2/L3Total
259MBcache
ECC-protected
cache arrays
More (much more) big & smart cache
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
© 2021 IBM Corporation29
LinuxONE
z15Chip
L3128MB
L3128MB
Open Standard
Compression Co-Processor
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
Core
CorezEDC Compressionand Asynch Data Mover
© 2021 IBM Corporation30
LinuxONE
L1/L2/L3Total
259MBcache
9.1BTransistors
12 Cores
ECC-protected
cache arrays
More (much more) big & smart cache
Compression Co-Processor
z15Chip
L3Chip
Cache128MB
L3Chip
Cache128MB
zEDC Compressionand Asynch Data Mover
© 2021 IBM Corporation33
LinuxONE
ProcessorDrawer
Chip Chip ChipChip
Up to 4 Chips(Sockets)
Per Drawer
© 2021 IBM Corporation34
LinuxONE
ProcessorDrawer
SCL4
960MBChip Chip ChipChip
SC / L49.7B
Transistors
More (lots more)Big Smart
Cache
StorageControl
Chip
© 2021 IBM Corporation35
LinuxONE
The 5th
DIMM is “RAIM”
backup for the other 4
ProcessorDrawer
SCL4
960MB
Memory DIMMs (5 of ‘em)
Memory DIMMs (5 of ‘em)
Chip Chip ChipChip
Up to 8TBUseable
RAIM RAM
RAIM: Redundant
Array of Independent
Memory
Failure-tolerant memory
subsystem
This is engineering!
(you don’t get to 50++ year MTBF without this
© 2021 IBM Corporation36
LinuxONE
PCIe Gen4
ProcessorDrawer
SCL4
960MB
PCIe Fanouts (Connectors) to other drawers
Memory DIMMs (5 of ‘em)
Memory DIMMs (5 of ‘em)
Chip Chip Chip Chip
FanoutsTo Sysplex Couplers
FanoutsTo Other
Processor or I/ODrawers
© 2021 IBM Corporation37
LinuxONE
Processor Drawer4 Chips
Per Drawer
SC / L49.7B
Transistors
FanoutsTo Sysplex Couplers
FanoutsTo Other
Processor and I/ODrawers
Motherboard
Up to 8TBUseable
RAIM RAM
More (lots more)Big Smart
Cache
ProcessorDrawer
StorageControl (SC)
and L4 Cache960MB
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
SMPFabricA-Bus
SMPFabricA-Bus
SMPFabricA-Bus
SMPFabricA-Bus
Memory DIMMs
Memory DIMMs
Memory DIMMs
Memory DIMMs
z15 ChipL3ChipCache128MB
L3ChipCache128MB
zEDC Compression
z15 ChipL3ChipCache128MB
L3ChipCache128MB
zEDC Compression
z15 ChipL3ChipCache128MB
L3ChipCache128MB
zEDC Compression
z15 ChipL3ChipCache128MB
L3ChipCache128MB
zEDC Compression
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
PCIeGen4
Fanout
Memory DIMMs
Memory DIMMs
Memory DIMMs
Memory DIMMs
Memory DIMMs
Memory DIMMs
© 2021 IBM Corporation38
LinuxONE
THE LINUXONE III CEC
Drawers to CECCEC: Central Electronics ComplexThe Processor Nest
© 2021 IBM Corporation40
LinuxONE
Processor Drawer
LT1 Model Drawers Cores Memory
(RAIM)Max34 1 1 to 34 Up to 8TB
Start at512GB RAIM
© 2021 IBM Corporation41
LinuxONE
LT1 Model Drawers Cores Memory
(RAIM)
Max71 2 1 to 71 Up to 16TB Processor Drawer
Processor Drawer
Fully Interconnected
SMP Fabric
© 2021 IBM Corporation42
LinuxONE
LT1 Model Drawers Cores Memory
(RAIM)
Max108 3 1 to 108 Up to 24TB
Processor Drawer
Processor Drawer
Processor Drawer
© 2021 IBM Corporation43
LinuxONE
LT1 Model Drawers Cores Memory
(RAIM)
Max145 4 1 to 145 Up to 32TB
Processor Drawer
Processor Drawer
Processor Drawer
Processor Drawer
© 2021 IBM Corporation44
LinuxONE
LT1 Model Drawers Cores Memory
(RAIM)
Max190 5 1 to 190 Up to 40TB
Processor Drawer
Processor Drawer
Processor Drawer
Processor Drawer
Processor Drawer
Scale to200,000
MIPS
© 2021 IBM Corporation45
LinuxONE
CEC: LinuxONE III LT1
LT1 Model Drawers Cores Memory
(RAIM)
Max34 1 1 to 34 Up to 8TB
Max71 2 1 to 71 Up to 16TB
Max108 3 1 to 108 Up to 24TB
Max145 4 1 to 145 Up to 32TB
Max190 5 1 to 190 Up to 40TB
1 to 190 Cores 512GB
to 40TBRAIM
Fully Interconnected
SMP Fabric
Scale to200,000
MIPS
© 2021 IBM Corporation47
LinuxONE
CEC: LinuxONE III LT2LT2
Model Drawers Cores Memory (RAIM)
Max04 1 1 to 4 Up to 2TB
Max13 1 1 to 13 Up to 4TB
Max21 1 1 to 21 Up to 4TB
Max31 1 1 to 31 Up to 8TB
Max65 2 1 to 65 Up to 16TB
1 to 65 Cores 256GB
to 16TBRAIM
Fully Interconnected
SMP Fabric
Max04 Max13 Max21 Max31
Max65
© 2021 IBM Corporation49
LinuxONE
Following…
§ Following LinuxONE III LT1 server rack-buildout scenarios• and several LinuxONE III LT2 rack scenarios
§ They are examples§ They illustrate key concepts§ There are many more
© 2021 IBM Corporation57
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
© 2021 IBM Corporation58
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Reserved
Reserved
To Assure Processing
Scale-ability, Reserve 1 or 2
Spaces for Future
Processor Drawers
Radiator Radiator
© 2021 IBM Corporation59
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
I/O Drawer1
Reserved
Reserved
Everything that happens next CAN BE non-disruptiveRadiator Radiator
© 2021 IBM Corporation60
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
I/O Drawer1
19” Rack Z
Reserved
Reserved
Radiator Radiator
© 2021 IBM Corporation61
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
I/O Drawer1
19” Rack Z
I/O Drawer2
Reserved
Reserved
Radiator Radiator
© 2021 IBM Corporation62
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
I/O Drawer1
19” Rack Z
I/O Drawer3
I/O Drawer2
Reserved
Reserved
Radiator Radiator
© 2021 IBM Corporation63
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
I/O Drawer1
19” Rack Z
I/O Drawer3
I/O Drawer2
Reserved
Radiator Radiator
© 2021 IBM Corporation64
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
I/O Drawer1
19” Rack Z
I/O Drawer6
I/O Drawer5
I/O Drawer4
I/O Drawer3
I/O Drawer2
Reserved
Radiator Radiator
© 2021 IBM Corporation65
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
I/O Drawer1
19” Rack Z
I/O Drawer6
I/O Drawer5
I/O Drawer4
I/O Drawer3
I/O Drawer2
19” Rack C
Reserved
Radiator Radiator
© 2021 IBM Corporation66
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
I/O Drawer1
19” Rack Z
I/O Drawer6
I/O Drawer5
I/O Drawer4
I/O Drawer3
I/O Drawer2
19” Rack C
I/O Drawer7
Reserved
Radiator Radiator
© 2021 IBM Corporation67
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
Processor Drawer 3
I/O Drawer1
19” Rack Z
I/O Drawer6
I/O Drawer5
I/O Drawer4
I/O Drawer3
I/O Drawer2
19” Rack C
I/O Drawer7 Radiator Radiator
© 2021 IBM Corporation68
LinuxONE
LinuxONE III LT1 Racks: Scenario 2 – iPDU Powered CPU Heavy
19” Rack A
Processor Drawer 1
Processor Drawer 2
Processor Drawer 3
I/O Drawer1
19” Rack Z
I/O Drawer6
I/O Drawer5
I/O Drawer4
I/O Drawer3
I/O Drawer2
19” Rack C
I/O Drawer11
I/O Drawer10
I/O Drawer9
I/O Drawer8
I/O Drawer7
Max’dOut
Config
Radiator Radiator
© 2021 IBM Corporation70
LinuxONE
Computer Science Core Performance
Programming Instructions
Memory Instructions Cycles Seconds
CycleInstructionsWorkloadWorkload+ **
CPU Time (Seconds)
Workload= Fastest
Thread
Superscalar
Clock SpeedBig Smart Cache
Modern Powerful
Instruction Set
© 2021 IBM Corporation71
LinuxONE
Computer Science Workload Performance
Programming Instructions
Memory Instructions Cycles Seconds
CycleInstructionsWorkloadWorkload+ **
CPU Time (Seconds)
Workload=
Synchronous I/O Wait in Seconds
Workload
Dispatch-able Wait Time in
Seconds
Workload+
Total Time (Seconds)
Workload=
CPU Time in Seconds
Workload+
Fastest Thread
Fastest Server
Fast I/O
Smart Software
Execution Threads
© 2021 IBM Corporation72
LinuxONE
Computer Science Core Performance
Programming Instructions
Memory Instructions Cycles Seconds
CycleInstructionsWorkloadWorkload+ **
CPU Time (Seconds)
Workload=
SIMD
Mega-Cache
Up to 12 Instructions Per Cycle
5.2GHz High
Frequency
Fastest Thread
Transactional Memory
LZ Compression
Smart Cache
BranchPredict
Pause-less GC Hiperdispatch
Packed Decimal
Out of Order Execution
Huffman Compression
New Compiler
Instructions
Quad TLLB
Superscalar
CPACF
zLibCompression Co-Processor
Modulo Arithmetic
Asynchronous Data Mover
SORT Accelerator
© 2021 IBM Corporation73
LinuxONE
Computer Science Workload Performance
Programming Instructions
Memory Instructions Cycles Seconds
CycleInstructionsWorkloadWorkload+ **
CPU Time (Seconds)
Workload=
Synchronous I/O Wait in Seconds
Workload
Dispatch-able Wait Time in
Seconds
Workload+
Total Time (Seconds)
Workload=
CPU Time in Seconds
Workload+
SMT2
SIMD Instructions
Mega-Cache
Up to 12 Instructions Per Cycle
5.2GHz High
Frequency
Fastest Thread
16GbpsFICON
Mega Memory
Transactional Memory
Fastest Server
WLM
LZ Compression
Smart Cache
BranchPredict
Pause-less GC
Compression
SMC-D SMC-R
zHyperlink 25GbpsOSA
Large SMPs
Hiperdispatch
Packed Decimal
Out of Order Execution
Huffman Compression
FlashExpress
Hyperswap
CF Duplexing
New Compiler
Instructions
Quad TLLB
Superscalar
CPACFEncryption
32GbpsFCP
IFLs
CPs
zIIPs
SAPs
IFP
SORTL
zLibCompression
© 2021 IBM Corporation74
LinuxONE
MIPS
§ Millions of Instructions Per Second
§ Mileage In-between Processing Systems
§ Meaningless Information Per Second
§ Meaningless Information Propagated by Sales-reps
© 2021 IBM Corporation75
LinuxONE
LSPR to the Rescue
§ https://www-01.ibm.com/servers/resourcelink/lib03060.nsf/pages/lsprindex?OpenDocument
§ Large Systems Performance Reference
• The IBM Large System Performance Reference (LSPR) ratios represent IBM's assessment
of relative processor capacity in an unconstrained environment for the specific benchmark
workloads and system control programs specified in the tables. Ratios are based on
measurements and analysis.
§ A Capacity Planning Resource
§ A Real(ish) set of Benchmarks for Mainframes
• Almost certainly the best set of benchmarks in the industry
§ zPCR – z Processor Capacity Reference – an Open Tool Full of LSPR Data & More
• https://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/PRS1381
© 2021 IBM Corporation79
LinuxONE
9 CMOS Generations of IBM ZCORES and CHIPS and Drawers
Core: 278 > 443 > 600 > 990 > 1280 > 1650 > 1906 > 2020 > 2232Chip: 278 > 876 > 2317 > 3701 > 4833 > 9097 > 13627 > 17783 > 23414Server: 3804 > 11,391 > 23,716 > 43,426 > 68,410 > 103,699 > 154,904 > 195,496 > 240,718
© 2021 IBM Corporation80
LinuxONE
IBM Z Chip Capacity Metrics
Look how much more work we can do!
Look how much more work we do per cycle!!!
© 2021 IBM Corporation81
LinuxONE
IBM Z Chip Technology Metrics
Look how much more energy
efficient we are!
© 2021 IBM Corporation82
LinuxONE
Moore’s Law
§ https://en.wikipedia.org/wiki/Moore%27s_law
§ Moore's law is the observation that the number of transistors in a dense integrated circuitdoubles about every two years. The observation is named after Gordon Moore, the co-founder of Fairchild Semiconductor and CEO of Intel, whose 1965 paper described a doubling every year in the number of components per integrated circuit[2] and projected this rate of growth would continue for at least another decade.[3] In 1975,[4] looking forward to the next decade,[5] he revised the forecast to doubling every two years.[6][7][8
© 2021 IBM Corporation83
LinuxONE
MOoRE ON MOORE’S LAW for MORE
§ n April 2005, Gordon Moore stated in an interview that the projection cannot be sustained indefinitely: "It can't continue forever. The nature of exponentials is that you push them out and eventually disaster happens." He also noted that transistors eventually would reach the limits of miniaturization at atomic levels:
§ In terms of size [of transistors] you can see that we're approaching the size of atoms which is a fundamental barrier, but it'll be two or three generations before we get that far—but that's as far out as we've ever been able to see. We have another 10 to 20 years before we reach a fundamental limit. By then they'll be able to make bigger chips and have transistor budgets in the billions.[100]
© 2021 IBM Corporation88
LinuxONE
Scale Out• Execute up to 1 trillion secure web transactions per
day on a LinuxONE III server • Scale-out to 2.4 million Docker containers in a
LinuxONE III• Consolidate 100 x86 running https to 1 LinuxONE III • 20% TCO reduction over 5 years
IBM z15 Chip and LinuxONE III Holistic Performance
Competitive Results• Up to 2.3x more throughput running pgBench 9.6.1
read-only benchmark on PostgreSQL 11.1 vs. compared x86
• Up to 2.1x more throughput running YCSB 0.15 (write-heavy) benchmark on MongoDB 4.0.6 vs. compared x86
• Up to 2.8x more throughput running Cassandra-stress benchmark on ScyllaDB 2.3.1 vs. compared x86 platform
• Up to 2.6x more throughput running DayTrader 7 benchmark on WAS Liberty vs. compared x86
Data Compression• Compress up to 275 GB web data per second on a
LinuxONE III server • Up to 2.7x lower latency and up to 1.8x less CPU at 2.6x
less network bandwidth running NGINX web server on LinuxONE III using Integrated zEDC to compress before encryption (up to 30x lower latency and up to 28x less CPU when compared to software compression)
• Up to 27x faster Db2 LUW database backup using 30x less CPU time using Integrated zEDC vs. compared x86 with software compression. (up to 20x faster using 23x less CPU time vs. LinuxONE Emperor II)
• Up to 6.5x faster MongoDB database dump using 5.5x less CPU time using Integrated zEDC vs. compared x86 with software compression. (up to 5.4x faster using 5.4x less CPU time vs. LinuxONE Emperor II)
• 3.2x faster MongoDB Dump on encrypted btrfs on LinuxONE III using Integrated Accelerator for zEDCcompression versus compared x86 with software compression
Instant Recovery• Start 100 KVM guests, in parallel, 2.1x faster
on LinuxONE III LPAR versus using a compared x86
IBM Private Cloud and SSC• Run 2.4x more AcmeAir containers per core
deployed on ICP on LinuxONE III vs. compared x86• Scale-up AcmeAir on SSC4ICP to 12 IFLs with 79%
scaling efficiency on a LinuxONE III LPAR
Note: Client results will varyPlease see approved claims for details and disclaimers
© 2021 IBM Corporation89
LinuxONE
FavorableEconomicOutcomes
Risk Mitigation
Operational Excellence
Fewer Cores – More Capacity
Flexibility
Resilience Built-In
Security without Compromise
Investment ProtectionFewer Cores – Cheaper Software
Performance
Throughput Scalability
The z15 Chip is Awesome – and so are LinuxONE III Servers!
© 2021 IBM Corporation90
LinuxONE
The Last Page
Questions?Comments?Requests?
THANK YOU for your Time