POWER7
Technology Insight
Wayne Huang
Fang Shu Xin
IBM CONFIDENTIAL – FOR IBM AND BP USE ONLY – NOT FOR DISCLOSURE TO CUSTOMERS
© 2010 IBM Corporation
IBM Power Systems
Agenda….
POWER7 Product Family
POWER7 Processor
Active Memory Expansion
POWER7 TPMD
OS Support on POWER7
POWER7 Servers
Power 750
Power 755
Power 770
Power 780
RAS Update
I/O Update
Summary
Power your planet
© 2010 IBM Corporation
IBM Power Systems
0
100
200
300
400
500
600
700
JS23 JS43 520 550 750 560 570/16 570/32 770 780 595
POWER7 System Highlights
Balance System Design
Cache, Memory, and IO
POWER7 Processor Technology
6 thImplementation of multi-core design
On chip L2 & L3 caches
POWER7 System Architecture
Blades to High End offerings
Enhances memory implementation
PCIe, SAS / SATA
Built in Virtualizatio n
Memory Expansion
VM Control
Green Technologies
Processor Nap & Sleep Mode
Memory Power Down support
Aggressive Power Save / Capping Modes
Availability
Processor Instruction Retry
Alternate Process Recovery
Concurrent Add & Services
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Power 2010 全新产品线
全新发布的Power7 产品 线
Pow er 750 (8, 16,32 C ore)
Pow er 755 ( 32 C or e) for HPC
Pow er 770 (12, 24,36,48 C ore)
Pow er 780 (16, 32,48,64 C ore)
POWER6 产品 线2010 年 继续支持
Pow er 520, Bl ades
Pow er 550
Pow er 560
Pow er 570
Pow er 575
Pow er 595
Power 750
Power 770
Power 755
Power 780
Power 7 Systems
Power 6 Systems
Power 520
Power Blades
Power your planet
Power 595
Power 570
Power 575
Power 560
Power 550
© 2010 IBM Corporation
IBM Power Systems
POWER7 Processor
POWER7
Pushing the
Limits
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Challenge: Beating Physics to Realize Multi-core Potential
Need to Amplify Effective
Socket Throughput
to Close Gap and
Achieve Potential
Compute Throughput Potential
Socket Throughput Limitation
Power your p lanet
(Physical signal economics)
© 2010 IBM Corporation
IBM Power Systems
Single Image Virtualized/Cloud
Trends in Server Evolution
Emerging Entry Server
Virtualiz ed/Cloud Platform
- A simple matter of riding
the multi-core trend?
- Ad d more core s to the die,
Enabled by:
8-core
8-core
- Technology
- Innovation
Driven by:
8-core
8-core
- IT Evolution
- Economics
Time
Traditi onal Entry Server
Single Image Platform
16 to 32-way SMP Server
2 to 4 socket
Traditional High-End Server
Virtualiz ed Consolidation Platform
beef up some interfaces,
and scale to a large SMP?
2-core 2-core
2-core 2-core
2 to 4 socket
4 to 8-way SMP Server
Powe r y our pl a ne t
8 to 32 socket
16 to 64-way SMP Server
* Statements regarding SMP servers
do not imply that IBM will introduce
a system with this capability.
© 2010 IBM Corporation
IBM Power Systems
Single Image Virtualized/Cloud
Trends in Server Evolution
Emerging Entry Server
Virtualiz ed/Cloud Platform
- A simple matter of riding
the multi-core trend?
- Ad d more core s to the die,
Enabled by:
- Technology
- Innovation
8-core
8-core
beef up some interfaces,
and scale to a large SMP?
Not so simple:
Driven by:
8-core
- IT Evolution
- Economics
Time
Traditi onal Entry Server
Single Image Platform
16 to 32-way SMP Ser v er
2 to 4 socket
8 - co re
- Emerging entry servers
have characteristics similar
to traditional high-end
large SMP servers
Traditional Hig h - End Server
Virtualiz ed Conso l i d ation Platform
Achieving solid virtual
machine performance
2-core 2-core
requires a Balanced
2-core 2-core
2 to 4 socket
4 to 8-way SMP Server
Powe r y our pl a ne t
8 to 32 socket
16 to 64-way SMP Server
System Structure.
* Statements regarding SMP servers
do not imply that IBM will introduce
a system with this capability.
© 2010 IBM Corporation
IBM Power Systems
Single Image Virtualized/Cloud UltraScale Cloud
Trends in Server Evolution
Enabled by:
Emerging Entry Server
Virtualiz ed/Cloud Platform
8-core
8-core
Emerging High-End Server
UltraScale Cloud Platform
- Technology
- Innovation
Driven by:
8-core
8-core
- IT Evolution
- Economics
Time
Traditi onal Entry Server
Single Image Platform
16 to 32-way SMP Server
2 to 4 socket
Traditional High-End Server
Virtualiz ed Consolidation Platform
8 to 32 socket
64 to 256-way SMP Server
Same enablers and
driving factors apply
at larger scale
2-core 2-core
2-core 2-core
2 to 4 socket
4 to 8-way SMP Server
Powe r y our pl a ne t
8 to 32 socket
16 to 64-way SMP Server
* Statements regarding SMP servers
do not imply that IBM will introduce
a system with this capability.
© 2010 IBM Corporation
IBM Power Systems
Challenge: How does POWER7 mai ntain the Bal ance?
Need to Amplify Effective
Socket Throughput
to Close Gap and
Achieve Potential
Compute Throughput Potential
Cache Hierarchy Technology
and Innovation
Socket Throughput Limitation
Power your p lanet
(Physical signal economics)
© 2010 IBM Corporation
IBM Power Systems
Challenge: How does POWER7 mai ntain the Bal ance?
Need to Amplify Effective
Socket Throughput
to Close Gap and
Achieve Potential
Compute Throughput Potential
Advances in Memory Subsystem
Cache Hierarchy Technology
and Innovation
Socket Throughput Limitation
Power your p lanet
(Physical signal economics)
© 2010 IBM Corporation
IBM Power Systems
Challenge: How does POWER7 mai ntain the Bal ance?
Need to Amplify Effective
Socket Throughput
to Close Gap and
Achieve Potential
Compute Throughput Potential
Advances in Off-Chip Signaling
Technology
Advances in Memory Subsystem
Cache Hierarchy Technology
and Innovation
Socket Throughput Limitation
Power your p lanet
(Physical signal economics)
© 2010 IBM Corporation
IBM Power Systems
Challenge: How does POWER7 mai ntain the Bal ance?
Need to Amplify Effective
Socket Throughput
to Close Gap and
Achieve Potential
Compute Throughput Potential
Exploit Long Ter m Investmen t
in Coherence Innovation
Advances in Off-Chip Signaling
Technology
Advances in Memory Subsystem
Cache Hierarchy Technology
and Innovation
Socket Throughput Limitation
Power your p lanet
(Physical signal economics)
© 2010 IBM Corporation
IBM Power Systems
POWER7 Processor
POWER7
Processor
Power your planet
© 2010 IBM Corporation
IBM Power Systems
20+ Years of POWER Processors
45nm
1.0um
POWER1
-AMERICA’s
Muskie A35
-Cobra A10
-64 bit
.72um
RSC
RS64IV Sstar
RS64III Pulsar
RS64II North Star
RS64I Apache
BiCMOS
.5um
.35um
POWER2
P2SC
.6um
-601
.5um
.35um
TM
-603
.5um
.35um
.25um
604e
.25um
.22um
POWER3
-630
.18um
TM
180nm
POWER4
-Dual Core
TM
65nm
130nm
TM
POWER5
-SMT
POWER6
-Ultra High Frequency
TM
Major POWER® Innovation
-1990 RISC Architecture
-1994 SMP
-1995 Out of Order Execution
-1996 64 Bit Enterprise Architecture
-1997 Hardware Multi-Threading
-2001 Dual Core Processors
-2001 Large System Scaling
-2001 Shared Caches
-2003 On Chip Memory Control
-2003 SMT
-2006 Ultra High Frequency
-2006 Dual Scope Coherence Mgmt
-2006 Decimal Float/VSX
-2006 Processor Recovery/Sparing
-2009 Balanced Multi-core Pr ocesso r
-2009 On Chip EDRAM
Next Gen.
POWER7
-Multi-core
1990 1995 2000 2005 2010
Power your planet
* Dates represent approximate proc essor power-on dates, not sys tem availability
© 2010 IBM Corporation
IBM Power Systems
IBM risc processors have many innovations..
pSeries p640,
64bit
P2,P3,P4
64bit
32bit
RS64
Apache
125
Power3
200+
RS64-II
Northstar
262.5
604e
332 / 375
1998
RS64-II
Northstar
340
1999 2000
p610
Power3-II
333 / 375 / 450
RS64-III
Pulsar
450
pSeries p620, p660, & p680
POWER4
1.1+GHz
RS64-IV
Sstar
600+ / 750
2001
POWER4
1.5GHz
Regatta
7450
800MHz/1.0GHz
2002
POWER4
1.8GHz
2003
Copper =
Power your planet
& SOI =
& low-k =
Σ
© 2010 IBM Corporation
IBM Power Systems
Processor Technology Roadmap
POWER6
65 nm
POWER5
130 nm
POWER4
180 nm
POWER8
POWER7
45 nm
Dual Core
Chip Multi Processing
Distributed Switch
Shared L2
Dynamic LPARs (32)
2001
Power your planet
Dual Core
Enhanced Scaling
SMT
Distributed Switch +
Core Parallelism +
FP Performance +
Memory bandwidth +
Virtualization
2004
Dual Core
High Frequencies
Virtualization +
Memory Subsystem +
Altivec
Instruction Retry
Dyn Energy Mgmt
SMT +
Protection Keys
2007
Multi Core
On-Chip eDRAM
Power Optimized Cores
Mem Subsystem ++
SMT++
Reliability +
VSM & VSX (AltiVec)
Protection Keys+
2010
Concept Phase
© 2010 IBM Corporation
IBM Power Systems
Processor Designs contrast: 0.278nm H2O
POWER5 POWER5+ POWER6 POWER7
Technology 130 nm 90 nm 60 nm 45 nm
Size 389 mm
2
245 mm
2
341 mm
2
567 mm
Transistors 276 M 276 M 790 M 1.2 B
Cores 2 2 2 4 / 6 / 8
Frequencies 1.65 GHz 1.9 GHz 3-5 GHz 3-4 GHz
L2 Cache 1.9 MB Shared 1.9 MB Shared 4 MB / Core 256 KB / Core
L3 Cache 36 MB 36 MB 32 MB 4 MB / Core
Memory Cntrl 1 1 2 / 1 2
LPAR 10 / Core 10 / Core 10 / Core 10 / Core
Power your planet
© 2010 IBM Corporation
2
IBM Power Systems
POWER6 / POWER7
POWER6
M
E
M
O
R
Y
L3
Mem
Ctrl
Alti
Vec
L3
Dir
SMT
Core
4MB
SMT
Core
4MB
L2
Bus Fabric Controller
GX Bus Cntrl
GX+ Bridge
Chip
to Chip
L2
to Chip
Alti
Vec
Chip
L3
Dir
Mem
Ctrl
L3
M
E
M
O
R
Y
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER6 / POWER7
POWER7
PO W ER 6
M
L3
E
M
O
R
Y
Power your planet
Mem
Ctrl
Alti
Vec
SMT
Core
SMT
Core
Alti
Vec
L3 Cache
L3
Dir
4MB
L2
Bus Fabric Controller
GX Bus Cntrl
GX+ Bridge
Chip
to Chip
eDRAM (Embedded D y namic RAM )
L3 — 6:1 latency improvement (vs. external L3) and 2x BW improvements
Capacitor vs transister
1/3 space (vs 6Trn SRAM cell), 1/5 standby power of standard SRAM
Soft error rated 250x lower than SRAM
Savings of ~ 1.5B transistors over other RAM
4MB
L2
Chip
to Chip
L3
Dir
Mem
Ctrl
L3
M
E
M
O
R
Y
© 2010 IBM Corporation
IBM Power Systems
POWER6 / POWER7
POWER7
M
E
M
O
R
Y
P
O
W
E
R
G
X
B
U
S
Mem
Ctrl
SMT
Core
A lti
L2 L2 L2 L2
L2 L2 L2 L2
Ve c
L3
Dir
4M B
L2 L2 L2 L2
L2 L2 L2 L2
SMT
Core
Bus Fa br i c Cont r ol l e r
Bus Fabric Controller
SMT
Core
SM T
Core
L3 Cache
L2
SMT
Core
SMT
Core
SM T
Core
4M B
L2
SMT
Core
A lti
Ve c
SMT
Core
L3
SMT
Dir
Core
Mem
Ctrl
GX Bus Cntrl
GX+ Bridge
Chip
to Chip
Footprints of working set:
Private footprints automatically migrates to fast local r eg i on 4m/cor e
Shared data automatically cloned to multiple private regions
Chip
to Chip
M
E
M
O
R
Y
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7 Processor Chip
Cores : 8 ( 4 / 6 core options )
Local SMP Links
POWER7
CORE
L2 Cache
L2 Cache
POWER7
CORE
POWER7
F
CORE
A
S
T
L2 Cache
L3 REGION
L3 Cache and
Chip Interconnect
L2 Cache
POWER7
CORE
Remote SMP & I/O Links
POWER7
CORE
L2 Cache
L2 Cache
POWER7
CORE
Binary Compatibility with
POWER6
POWER7
CORE
L2 Cache
L2 Cache
POWER7
CORE
567mm
2
Technology:
45nm li’thography, Cu, SOI, eDRAM
Transistors: 1.2 B
Equivalent function of 2.7B
eDRAM efficiency
Eight processor cores
MC1 MC0
12 execution units per core
4 Way SMT per core – up to 4 threads per core
32 Threads per chip
L1: 32 KB I Cache / 32 KB D Cache
L2: 256 KB per core
L3: Shared 32MB on chip eDRAM
Dual DDR3 Memory Controllers
90 GB/s Memory bandwidth per chip
Scalability up to 32 Sockets
360 GB/s SMP bandwidth/chip
20,000 coherent operations in flight
Power your planet
4th Generation SMP Fabric Bus
3rd Generation Multi-Threading
Energy Optimized Desig n
Enhanced GX System Buses
DDR3 memory
On-chip eDRAM L3
© 2010 IBM Corporation
IBM Power Systems
Cache Hierarchy Technology a nd Innovat ion
Local SMP Links Remote SMP + I/O Links
Core
L2 Cache
Mem Ctrl Mem Ctrl
L2 Cache
Core
L3 Cache and Chip Interconnect
Core
L2 Cache
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Cache Hierarchy Technology a nd Innovat ion
Local SMP Links Remote SMP + I/O Links
Core
L2 Cache
Mem Ctrl Mem Ctrl
L3 Cache and Chip Interconnect
L2 Cache
Core
Core
L2 Cache
Fast Local
L3 Region
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7 Core
64-bit PowerPC architecture v2.07
Execution Unit s
• 2 Fixed Point Units
• 2 Load Store Units
• 4 Double Precision Float ing Point Units
• 1 Branch
• 1 Condition Register
• 1 Vector Unit
• 1 Decimal Floating Point Uni t
• 6 Wide Dispatch
• Units include distributed Recovery Function
Out of Order Execution excellent for commercial load
ISU
IFU
CRU/BRU
DFU
FXU
L2 Cache
VSX
FPU
LSU
Modes: POWER6, POWER6+ and POWER7
e.g. partition mobility+
POWER7 continue s to s u pport VMX / Extends SIMD supp or t with VSX
2 VSX (Vec Scalar Ext) units that can each handle 2 Double-Precision FP instructions
8 FLOPS per cycles
VSX units can also handle 4 Single Precision instructions per cycle
VSX instruction set support for vector and scalar instructions
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Advances in Memory Subsystem
Memory Subsystem Requiremen t
for POWER Servers
Core
Need 10 to 20 GB/s
Sustained bandwidth
per Core
Need 16 to 32 GB
of Storage per Core
Challenge
for Multi-core POWER7
Socket Challenge:
4x growth in memory bandwidth
and capacity needed per socket.
System Challenge:
Packaging more memory into
similar volume with
similar energy and cooling
constraints.
Energy Constraints
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Advances in Memory Subsystem
Multi-faceted Solution
POWER7 Chip
Memory
Controller
Memory
Controller
Advanced
Buffer
Chip
1) Dual Integrated DDR3 Controllers
- Massive 16KB scheduling window
per POWER7 chip insures high
channel and DIMM utilization
- Sparse access acceleration
- Advanced Energy Management
- Numerous RAS advances
2) Eight high speed 6.4 GHz channels
- New low power differential signaling
- Sustained 100+ GB/s per socket
3) New DDR3 buffer chip architecture
- Larger capacity support (32 GB / core)
Power your planet
- Energy Management support
- RAS enablementh
4) DDR3 DRAMs
- Supports 800, 1066, 1333, and 1600
* Statements regarding memory subsystem features do not imply that IBM will introduce a system with these capabili ties.
© 2010 IBM Corporation
IBM Power Systems
Memory Channel Bandwidth Evolution
POWER5
Memory Performance:
2x DIMM
POWER6
Memory Performance:
4x DIMM
POWER7
Memory Performance:
6x DIMM
D
D
D
D
R
R
3
3
D
D
D
D
R
R
3
3
D
D
D
D
R
R
3
3
D
D
D
R
3
D
D
D
R
R
3
3
D
D
R
3
DDR2 @ 553 MHz
Effective Bandwidth:
1.1 GB/s
Power your planet
DDR2 @ 553 / 667 MHz
Effective Bandwidth:
2.6 GB/sec
DDR3 @ 1066 MHz
Effective Bandwidth:
6.4 GB/sec
© 2010 IBM Corporation
IBM Power Systems
Exploit Long Term Investment in Coherence Innovation
Local SMP Links Remote SMP + I/O Links
Core
L2 Cache
Mem Ctrl Mem Ctrl
L2 Cache
Core
L3 Cache and Chip Interconnect
Core
L2 Cache
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Core
L2 Cache
L2 Cache
Core
Using local and remote SMP links, up to 32 POWER7 chips are connected
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Exploit Long Term Investment in Coherence Innovation
Up to 32 POWER7 chips form a massive SMP system.
Power your planet
* Statements regarding SMP servers
do not imply that IBM will introduce
a system with this capability.
© 2010 IBM Corporation
IBM Power Systems
Multi-threading Evolut ion
S80 HMT
Single thread Out of Order
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
POWER5 2 Way SMT
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
S80 Hardware Multi-thread
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
POWER7 4 Way SMT
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
No Thread Executing
Thread 2
Executing
Power your planet
Thread 0
Executing
Thread 3
Executing
Thread 1
Executing
© 2010 IBM Corporation
IBM Power Systems
0
0.5
1
1.5
2
2.5
SMT4 SMT2 Single
0
0.5
1
1.5
2
2.5
SMT4 SMT2 Single
IBM Confidential
POWER7 Multi-threading Opti ons
TurboCore option
50% of the cores active
MaxCore option
All cores active
Power your planet
Based of rPerf workload
© 2010 IBM Corporation
IBM Power Systems
0
0.5
1
1.5
2
SMT1 SMT2 SMT4
POWER7 SMT4
Requires POWE R 7 Mode
POWER6 Mode supports SMT1 and SMT2
Operating Sy st em Support
AIX 6.1 and AIX 7.1
IBM i 6.1 and 7.1
Linux
Dynamic Runtim e SM T scheduling
Spread work among cores to execute in
Standard Cache Option
All cores active
appropriate threaded mode
Can dynamical shift between modes as required:
SMT1 / SMT2 / SMT4
LPAR-wide SMT controls
ST, SMT2, SMT4 modes
smtctl / mpstat commands
Mixed SMT mode s su pported within sam e
LPAR
Requires use of “Resource Groups”
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7 TurboCore Mode
Power 780 TurboCore Chip
TurboCore Chips: 4 available cores
Aggregation of L3 Caches of unused cores.
TurboCore chips have a 2X the L3 Cache
per Chip available
4 TurboCore Chips L3 = 32 MB
Performance gain over POWER6.
Provi des up to 1.5X per core to core
Chips run at higher frequency:
Power reduction of unused cores.
With “Reboot”, System can be reconfigur ed
to 8 core mode.
ASM Menus
CoreL2Core
P
O
W
E
R
G
X
B
U
S
L2
Core
POWER7 Chip
CoreL2Core
L2
32 MB
L3 Cache
L2
Core
L2
Core
L2
L2
Core
S
M
P
F
A
B
R
I
C
TurboCores
Power your planet
Memory Interface
Unused
Core
© 2010 IBM Corporation
IBM Power Systems
TurboCore - Example
Single Node Power 780 system (maximum throughput)
One processor feature #4982 (0 of 16)
Two POWER7 processors
64 MB internal L3 cache
16 POWER7 cores @ 3.8 GHz
Up to 16 CoD processor core activation features #5469
Best total system capacity config urat io n
Single Node Power 780 system (TurboCore mode)
One processor feature #4982 (0 of 16)
Two POWER7 processors
64 MB internal L3 cache
System is configured for TurboCore mode
8 POWER7 cores @ 4.1 GHz available
Up to 8 CoD processor core activation features #5469
Up to 22% greater performance per core
Power your planet
X
X
TurboCore
Cores
X
X
X X
X X
Unused
Core
© 2010 IBM Corporation
IBM Power Systems
POWER7 Processor
POWER7
Active Memory
Expansion
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Active Memory Expansion
True
Memory
True
Memory
True
Memory
Expanded
Memory
Expanded
Memory
Expanded
Memory
Effectively up
to 100% more
True
Memory
True
Memory
True
Memory
Expanded
Memory
POWER7 Advantage
Expand memory beyond physical limits
More effective server consolidation
Run more application workload / users per partition
Run more partitions and more workload per server
The configuration is on a per-LPAR basis. When AME is enabled for a LPAR, the OS will compress a portion of the LPAR’ s memory and leave the remaining portion of m emory
uncompressed.
This results in memory effectively being broken up into two pools – a compressed pool and an uncompressed pool.
The operating system will dynamically vary the amount of memory that is compressed based on the workload and the configuration of the LPAR.
The operating system will move data between the compressed and uncompressed mem ory pools based on the memory access patterns of apps.
When an application needs to access data that is compressed, the operating sys tem will automatically decompress the data and move it from the compressed pool to the uncompressed
pool, making it available to the appl i cation.
When the uncompressed pool is full, the operating system will compress data and move it from the uncompressed pool to the compressed pool. This compression and decompression
activity is transparent to the application.
Because Active Memory Expansion relies on memory compression, some additional CPU utilization will be consumed when Active Memory Expansion is in-use. The amount of additional
CPU utilization needed for Active Memory Expansion will vary based on the workload and the level of memory expansion being used.
Expanded
Memory
Expanded
Memory
memory
Power your planet
© 2010 IBM Corporation
IBM Power Systems
#10
#9
#8
#7
#6
#5
#4
#3
#2
#1
Active Memory Expansion & Active Memory Sharing
Active Memory Expansion
Effectively gives more memory
capacity to the partition using
compression / decompression of
the contents in true memor y
AIX partitions only
Active Memory Sharing
Moves memory from one partition
to another
Best fit when one partition is not
busy when another partition is
busy
AIX, IBM i, and Linux partitions
Active Memory Expansion Active Memory Sharing
Supported, potentially a very nice option
Considerations
Only AIX partitions using Active Memory Expansion
Active Memory Expansion value is dependent upon com pressi bilit y of
data and available CPU resource
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Active Memory Expansion
Innovative POWER7 technology
For AIX 6.1 or later
For POWER7 servers
Uses compression/decompression to effectively expand the true physical
memory available for client workloads
Often a small amount of processor resource provides a significant incr ease in
the effective memory maximum
Processor resource part of AIX partition’s resource and licensing
Actual expansion results dependent upon how “com pressi ble” t he data being
used in the application
A SAP ERP sample workload shows up to 100% expansion,
Your results will vary
Estimator tool and free trial available
Will not compress AIX kernel
Will not compress pinned memory pages (pinned pages to see)
Will not compress File memory pages incl. code, cache & mmap (numperm to see)
Already compressed data (e.g. compress database objects) will not compress well
Computational memory good target (data:heap/stack, but not the code)
Localized memory access pattern works well with AME
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Excellent Compression Targets
Data only used on program initialization
Pages allocated but unused = full of zeros/blanks
Pages with lots of repeat data like database r eco r ds
Access Pattern
Some hot pages, some warm , some freezing
All pages equally used (HPC) – not so good
How can I work that out?
Power your planet
© 2010 IBM Corporation
IBM Power Systems
What is your Plan?
10GB
Memory Shrinking
for supporting more lpars
or workloads
Looks like 10GB but
is actually 8GB
Memory Growing
for constrained lpar
10GB
But want 14GB to
improve performance
2GB released
for another LPAR
Power your planet
Actually still using 10 GB but
looks like 14 GB
© 2010 IBM Corporation
IBM Power Systems
Permanent Enablement - Chargeable
One feature per server
No matter how many partitions use it
Permanent enablem ent with new server or v ia MES order
Enablement “V E T ” code applied to the VP D anchor card
Once enabled: no mechanism to move it to a different server
Power 750 & Power 755
#4792 AME E nabl em ent Feature
Power 770 & Power 780
#4791 AME E nabl em ent Feature
One-time, 60-day trial - No charge
Request via Capacity on Demand Web page
www.ibm.com/systems/power/hardware/cod/
Power your planet
© 2010 I B M C o r p o r at i o n
IBM Power Systems
Active Memory Expansion – Client Deployment
1
Planning Tool
A. Part of AIX 6.1 TL4
B.
Calculates data
compressibility
estimates CPU
overhead due to
Active Memory
Expansion
C. Provides initial
recommendations
Estimated Results
&
2
60-Day Tri al
A. One-time, temporarily
enablement
B. Config LPAR based
on planning tool
C. Use AIX tools to
monitor Act Mem Exp
environment
D. Tune based on actual
results
Actual Results
3
Deploy into Product ion
A. Permanently enable
Active Memory
Expansion
B. Deploy workload
into production
C. Continue to monitor
workload using AIX
performance tools
CPU Utilization
Memory Expansion
Power your planet
CPU Utilization
Memory Expansion
App. Performance
Memory Expansion
Performance
Time
© 2010 IBM Corporation
IBM Power Systems
Active Memory Expansion - Planning Tool
amepat -- AME Planning and Advisory Tool
Active Memory Expansion Modeled Statistics:
----------------------Modeled Expanded Memory Size : 8.00 GB
This sample partition
Expansion True Memory Modeled Memory CPU Usage
Factor Modeled Size Gain Estimate
--------- -------------- ----------------- -----------
1.21 6.75 GB 1.25 GB [ 19%] 0.00
1.31 6.25 GB 1.75 GB [ 28%] 0.20
1.41 5.75 GB 2.25 GB [ 39%] 0.35
1.51 5.50 GB 2.50 GB[ 45%] 0.58
1.61 5.00 GB 3.00 GB [ 60%] 1.46
Active Memory Expansion Recommendation:
--------------------The recommended AME configuration for this workload is to configure
the LPAR with a memory size of 5.50 GB and to configure a memory
expansion factor of 1.51. This will result in a memory expansion of
45% from the LPAR's current memory size. With this configuration,
the estimated CPU usage due to Active Memory Expansion is
approximately 0.58 physical processors, and the estimated overall
peak CPU resource required for the LPAR is 3.72 physical processors.
has fairly good
expansion potential
A nice “sweet” spot for
this partition appears
to be 45% expansion
• 2.5 GB gained memory
• Usi ng about 0.58 cores
additional CPU
resource
Tool included in AIX 6.1 TL4 SP2
Run tool in the partition of interest for memory expansion.
Input desired expanded memory size. Tool outputs different real memory
and CPU resource comb inations to achieve the desired effective memory.
Power your planet
© 2010 IBM Corporation
IBM Power Systems
How do we switch AME on?
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Is the machine AME Capable?
P ow e r y our pl a ne t
HMC
Server Properties
Capabilities
then scroll to the bottom
© 2010 IBM Corporation
IBM Power Systems
Activate on AME on the LPAR profi le
Hard reboot (not restart ) to
activate LPAR in AME mode
Expansion Factor:
Power your planet
1.0 = AM E on but inactive
1.2 to 1.5 = Good start point
10.0 = suicidal !
© 2010 IBM Corporation
IBM Power Systems
Dynamically changing the Expansion Factor
Use Dynamic LP AR Memory Add/Remove
and change the Expansion Factor
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Monitoring Active Memory Expansion in use
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7
TPMD
Power your planet
© 2010 IBM Corporation
IBM Power Systems
TPMD: Thermal Power Manageme nt Device
TPMD card is part of the base hardware confi guration.
Residing on the processor planar
TPMD function is comprised of a risk processor and data acquisition
TPMD monitor power usage and temperature s in real time
Responsible for thermal protection of the processor cards
Can adjust t he processor power and performance in real time.
If the temperature ex ceeds an upper (f unctional) threshold, TPMD
actively reduces power consumption by reducing processor voltage
and frequency or throttling memory as needed.
If the t emperat ure is lower than upper (functional) threshold, TPMD
will allows POWER7 cores to “Over clock
” if workloads demands are
present.
~10% up clock speed, beneficial for cpu intensive workload
Power your planet
© 2010 IBM Corporation
IBM Power Systems
3
3.2
3.4
3.6
3.8
4
4.2
4.4
POWER7 “Over Clock” Uplift
Power your planet
© 2010 IBM Corporation
IBM Power Systems
EnergyScale
EnergyScale is IBM Trademark. It consists of a built-in Thermal Power
Management Device (TPMD ) card and Power Executive software.
IBM Systems Director is also required to manage Energy -Scale functions.
EnergyScale is used to dynamically optimizes the processor performance
versus processor power and system workload.
IBM Systems Director is also required to manage AEM functions and
supports the following functions:
• Power Trending
• Thermal Reporting
• Static Energy Saver Mode
• Dynamic Energy Saver Mode
• Energy Capping
• Soft Energy Capping
• Processor Nap
• Energy Optimized Fan Control
• Altitude Input
• Processor Folding
Power your planet
© 2010 IBM Corporation
53
IBM Power Systems
What is ENERGY STAR?
ENERGY STAR is a program developed by the U.S.
Environmental Protection Agency (EPA) to reduce
energy consumption
Voluntary labeling program designed to identify and
promote energy efficient products
Computer servers that earn EPA’s ENERGY STAR
include:
Power 750 Express
Efficient power supplies that have smaller conversion losses and
generate less waste heat,
Capabilities to measure real time power use
Advanced power management features
Power and Performance Data Sheet
Power 750 Express and Power 755 are the first RISC
or Itanium ENERGY STAR-qualified server
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Power W orkload
Optimization
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Power Workload Optimization – the new features
Power Systems offers b alanced systems desig ns
that automatically optimize workload performance
and capacity at either a system or VM level
✓ TurboCore™ for max per core performance for databases
✓ MaxCore for incredible parallelization and high capacity
✓ Intelligent Threads utilize more threads when workloads benefit
✓ Intelligent Cache technology optimizes cache utilization flowing it from core to core
✓ Intelligent Energy Optimization maximizes performance when thermal conditions allow
✓ Active Memory™ Expansion provides more memory for SAP
✓ Solid State Drives optimize high I/O access applications
Workload-Optimizing Features make POWER7
#1 in Transaction and Throughput Computing
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7
OS Support
Power your planet
© 2010 IBM Corporation
IBM Power Systems
58
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER7 based Systems provide two modes for each LPAR:
POWER6 & POWER6+ ,POWER7
P6 MODE (and P6+ )* P7 MODE Customer Value
2-Thread SMT 4-Thread SMT
8 Storage Keys
16 Storage Keys in P6+
VMX (Vector Multimedia
Extension / AltiV ec)
Affinity OFF by Default On by Default
64-core/128-thread Scaling 32-core / 128-thread Scaling
EnergyScale CPU Idle EnergyScale CPU Idle and Folding
32 Storage Keys
VSX (Vector Scalar Extension)
3-tier Memory , MicroPartition Affinity
64-core / 256-thread Scaling
256-core / 1024-thread Scaling
with NAP and SLEEP
吐能力、CPU 利用率
RAS、数据隔离的颗粒度、提升应用
键的使用
高性能计算
提升系统性能
高可扩展、服务器整合
提升能源效率
N/A Active Memory Expansion
Power your planet
扩展内存容量
© 2010 IBM Corporation
IBM Power Systems
OS Support on POWER7: IBM i ,AIX and Linux
Max Cores
IBM i
Release
POWER6 Mode POWER7 Mode
IBM i 6.1 32 / 64 32 / 128
Special Support 64/128 32 / 128
IBM i 7.1 32 / 64 32 / 128
Special Support 64/128 64 / 256
& Threads Supported
Linux
Max Cores
AIX
Release/TL
POWER6 Mode POWER7 Mode
AIX 5.3
(All TL9,10,11,12)
AIX 6.1 TL2, TL3 64 / 128 N/A
AIX 6.1 TL4 64 / 128 64 / 256
AIX 6.1 TL5 64 / 128 64 / 256
AIX 7.1 64 / 128 256 / 1024
& Threads Supported
64 / 128 N/A
Max Processors
& Threads Supported
POWER6 Mode POWER7 Mode
RHEL 5 ( Updates new er than U4) 64 / 128 N/A
SLES 10 SP# and newer 64 / 128 N/A
SLES 11 ( All Service Packs) 64 / 128 256 / 1024
RHEL 6 (Next major RHEL version) 64 / 128 256 / 1024
Power your planet
© 2010 IBM Corporation
IBM Power Systems
2009 – 20 11 AIX TL Roadmap
AIX 5.3
TL8
TL9
TL9
TL1 0
TL10
AIX 6.1
TL1
TL2
TL2
TL3
TL3
TL1 1
TL11
TL4
TL4
SP
SP
04/2010 10/2010
POWER7 Hardware Support
SP
SP
TL1 2
TL12
SP
SP
TL5
TL5
TL6
04/2011 10/2011 10/2009
SP
Service Pack
POWER7 Support
Power your planet
TL7
TL8
TL0
TL1
AIX 7.1
TL2
© 2010 IBM Corporation
IBM Power Systems
AIX support for POWER7
AIX supports Power 750, 755, 770, and 780 with P OWER7
processors offering more performance, energy
efficiency and scalab il ity
Initial AIX Levels supported
AIX 6 Technology Level 4 SP3 (GA 2/12)
AIX 5.3 Technology Level 11 SP2 (GA 3/16)
April AIX Technology Levels
AIX 6 Technology Level 5 (GA 4/23)
AIX 5.3 Technology Level 12 (GA 4/23)
Prior Technology levels
AIX 6 Technology Levels 2 and 3 (June 2010)
AIX 5.3 Technology Levels 9 and 10 (May 2010)
Power your planet
© 2010 IBM Corporation
IBM Power Systems
POWER Virtualization Features
DLP A R Processor add/remove
DLPAR I/O adapter add/remove
DLPAR Memory add
DLPAR Memory remove
Micro Partitioning
Capacity Upgrade on Demand
Virtual I/O Server
Integrated Virtualization Manager
Virtual SCSI (VIO Server)
Virtual Ether net ( VIO Serv er )
Workload Partitions (AIX only)
Application Mobility (AIX only)
Lx86 (Statement of direction for POWER7)
Shared Dedicated Capacit y (POWER6)
AIX
5.3
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes No No Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
No Yes No No No No
No Yes No No No No
No No No Yes Yes Yes
Yes Yes Yes Yes Yes Yes
AIX
6.1
IBM i
6.1.1
RHEL
5.4
SLES
V10 SP3
SLES
V11
Multiple Processor Pools (POWER6)
Live Partition Mobility (POWER6)
N_Port ID Virtualization (POWER6)
Active Memory Expansion (POWER7)
Active Memory Sharing (POWER6)
Integrated Virtual Ethernet (POWER6)
Activ e Memory Expansion (POWER7 / AIX 6.1)
Note: Some features are supported with specific fix levels or only on specific hardware models. Check the appropriate documentation.
Power your planet
Yes Yes Yes Yes Yes Yes
Yes Yes No Yes Yes Yes
Yes Yes Yes Yes No Yes
No Yes No No No No
Yes Yes Yes No No Yes
Yes Yes Yes Yes Yes Yes
No Yes No No No No
© 2010 IBM Corporation
IBM Power Systems
Partition Mobility
POWER6
POWER6+
POWER7
Binary Compatibility between POWER6 and POWER7
Leverage POWER6 / POWER6+ Compatibility Mode
Migrate partitions between POWER6 and POWER7 Servers
Forward and Backward
Power your planet
© 2010 IBM Corporation
IBM Power Systems
65
Power your planet
© 2010 IBM Corporation