POWER8. Open Innovation to Put Data to Work for SAP · PDF filePOWER8. Open Innovation to Put Data to Work for SAP ... SAP HANA Deployment Options for Different Use Cases ... Open

POWER8. Open Innovation to Put Data to Work for SAP

© 2015 IBM Corporation

Alfred FreudenbergerIBM North AmericaPower SystemsSAP on POWER Sales Leader

[email protected]

Blog: saponpower.wordpress.com

© 2015 International Business Machines Corporation

IBM Watson

Linux Centers around the world

Hybrid Cloud

OpenPOWER Foundation

3+ years, $2.5 billion R&D investment

Three major Linux distros

$1 billion

POWER on SoftLayer

Our POWER transformation journey

Aligned to the New Era (Cloud, Data, Open)

New Data-Centric Systems

Most Open Server in the Market

Simplified POWER8 Portfolio with Scale-Out and Enterprise (scale up)

Expanding Linux Ecosystem with 1000s of ISVs, 100,000s of Open Source Packages, and Dozens of Solutions

Growing University Program, with hundreds of schools

Double-digit Growth in CAMSS, Scale-Out Servers and OpenPOWER

2


© 2014 OpenPOWER Foundation© 2014 OpenPOWER Foundation

Fueling an Open Development Community

Complete member list at www.openpowerfoundation.org

Implementation / HPC / Research

Software

System / Integration

I/O / Storage / Acceleration

Boards / Systems

Chip / SOC

> 140


Announcing the IBM Power Systems S812LCoptimized for entry and small Hadoop workloads

� 1-socket, 2U

� Up to 10 cores (2.9-3.3Ghz)

� 1 TB Memory (32 DIMMs)

� 115GB/sec memory bandwidth

� 14 LFF (HDD/SSD) 84TB storage

� 4 PCIe slots, 2 CAPI enabled

� Default 3 year 9x5 warranty, 100% CRU

Improve the agility and reduce the cost of running Spark and Hadoop workloads

• 16X the memory capacity of Xeon E3 servers, 2X 1P Xeon E5 servers

• Complete the same Spark workloads for <½ the cost of Intel Xeon E5-2690 v3 systems

• 2.3X BETTER performance per dollar spent

• 94% more Spark workloads in the same rack space as Intel Xeon E5-2690 v3 systems

• 1.94X BETTER performance per system (10 core S812LC vs 24 core DL380)


Announcing the Power Systems S822LC for commercial

computing ideal for data in the cloud

� 2-socket, 2U

� Up to 20 cores (2.9-3.3Ghz)

� 1 TB Memory (32 DIMMs)

� 230GB/sec memory bandwidth

� 2x SFF (HDD/SSD), 2 TB storage

� 5 PCIe slots, 4 CAPI enabled, IB Add-in

� Default 3 year 9x5 warranty, 100% CRU

Simplify and optimize your infrastructure• 2.7X transactions/second/core with PostgreSQL vs Xeon E5-2699

v3-based servers

• 40% better price/performance with PostgreSQL vs Xeon E5-2699 v3-based servers

• Ideal economics for on-premises and in cloud applications


Speed insights in the cloud with the most secure, converged infrastructure with open choice – IBM PurePower System

Delivering innovation and choice

across all three OS – AIX, IBM i, Linux

Integrated Infrastructure Manager –

Enhanced RAS with management node failover via Virtual HMC (vHMC)


POWER8 RAS Feature Differentiation

7

RAS ItemLinux

Cluster

Linux Only

S8xyL

Scale Out

S8xy

Enterprise

E850

Enterprise

E870/E880

Redun / Hot Swap Power Supplies ◙ ● ● ● ●

Redundant Voltage Regulators ▬ ▬ ▬ ● ●

Integrated Spare Voltage Regs ▬ ▬ ▬ ● ●

OCC error handling w/ power safe mode ● ● ● ● ●

EPOW reporting ● ● ● ● ●

TOD Battery Concurrent Replace ▬ ▬ ▬ ● ●

Redundant Fans/Blowers ● ● ● ● ●

Hot Swap Fans / Blowers ● ● ● ● Upper Domain ●

Integrated Spare Fans/Blowers ▬ ▬ ▬ ● Lower Domain N/A

Hot Swap DASD / Media ● (DASD only) ● ● ● ●

Dual disk controllers (split backplane) ▬ ● ● ● N/A

Hot Swap PCI Adapter ▬ ● ● ● ●

EEH Support Partial Partial ● ● ●

Concurrent Op Panel Repair N/A ● ● ● ●

Triple redundant Ambient Thermal Sensors ▬ ▬ ▬ ● ●

Redundant Service Processors ▬ ▬ ▬ ▬ ●

Dynamic Srvc Processor Failover ▬ ▬ ▬ ▬ ●

Auto Service Proc Reset/Reload ▬ ● ● ● ●

Redundant System Clocks ▬ ▬ ▬ ▬ ●

Dynamic System Clock Failover ▬ ▬ ▬ ▬ ●

Redundant VPD ▬ ● ● ● ●

● Standard

◙ Optional

▬ Not Available


PowerVC: Increase IT productivity and agility � Advanced Policy Management: CPU & Memory capacity, host grouping and rules for VM system

affinity; Dynamic Resource Optimization with COD or Live Partition Mobility

� High Availability Improvements – Support for redundant HMCs, VM Restart, SVC Mirroring configurations and support for shared storage typically used by clusters

� Improved storage capabilities – Support for EMC Powerpath, IBM SDDPCM Multi-Path managers, Multi-volume VM capture and deploy, DS8000 NPIV and vSCSI storage

PowerVM: Virtualization without Limits� Expand application choice with support for Little Endian Linux Guest VMs� Novalink: Direct OpenStack connection to PowerVM Host to improve scalability *

� VM Mobility enhancements with SRIOV, resiliency improvements *

� Storage Tiers *

PowerVP Performance Monitoring for PowerVM� Export of PowerVP performance data to an external repository� Integration with the VIOS performance advisor� New thresholds and alerts� Little Endian guest support� Support for running monitor in a browser

PowerKVM: Open virtualization for scale out Linux systems� Host runs in Little Endian *

� Dynamically add vCPU and memory resources *

� SRIOV enhancements: PCI pass thru and Hot plug support *

� Dynamic Microthreading *

Power System SoftwareAn intelligent IT infrastructure for Cloud, Big Data, Analytics & Mobile


Delivering innovation and open technology with OpenStack across the entire Power Systems line

• Simplify and automate workload management across AIX, IBM i, and Linux for simpler and more scalable cloud deployments with:

– PowerVM NovaLink enabling direct OpenStackconnection to PowerVM host

– PowerVC Dynamic Resource Optimizer to actively balance workloads and resources based on business policy


• PowerVM NovaLink Architecture

− Improves Cloud scalability

− Allows direct OpenStack connection to PowerVM host and enables OpenStack community adoption of PowerVM drivers

− Simplified bare-metal PowerVM install

• vNIC support for SRIOV adapters enabling VM mobility and improved performance

• Live Partition Mobility Improvements

– Increased LPM resiliency when single VIOS fails and better NPIV verification

– Improved performance and network efficiency

• Shared Storage Pools Improvements

– Enhanced control of performance and data segregation with support of up to 10 Storage Tiers in a pool

– Ability to Grow a Virtual Disk

• Scaling to up to 5,000 VMs – Support for Managing PowerVM hosts using NovaLink facility (up to 5,000 VMs, 200 hosts and 500 VMs per host with PowerVC)

• Advanced Policy-based Management with Dynamic Resource Optimizer

– New policy control that actively balances workloads and resources for a pool of systems based on CPU usage

– Balances workloads by either moving workloads with Live Partition Mobility or by using Mobile Capacity on Demand (COD) to move cores to a workload with no movement of the workload

• New Storage Capabilities

– Support for DS8K NPIV and vSCSI storage,

• Support for PowerVM Multiple Shared Processor Pools during initial placing and migration of VMs

PowerVM v2.2.4Virtualization without limits

PowerVC v1.3Increasing productivity and agility

Power System Software

Delivering superior Cloud economics through increased scalability of the cloud infrastructure, more efficient management of resources and policy-based automation


• AIX Live Kernel Update: Non-disruptive OS update. Initial implementation for i-fixes (future plans for SPs and TLs and for commands and Libraries)

• Server side Flash Cache

–Storage agnostic Flash Cache

–Dedicated or virtualized implementation

• Install enhancements for stronger networking security by default

• PowerSC NERC profile

• Improved economics by automatically moving system and software entitlements during failover

− Minimize cost of redundant resource for HA and DR

• Minimize administration costs and errors through automation

– Automate PowerHA cluster alignment with Live Partition Mobility and Live Kernel update

– Cluster verification checks for AIX and PowerHA consistency

• Seamless cluster maintenance

– Upgrade from PowerHA V7.1.3 and beyond without disruption

– Automated failed repository disk replace

• Quarantine protection from failing, but not failed node

AIX 7.2Leadership performance,

resilience and security

PowerHA SystemMirror v7.2Resiliency without Downtime

AIX … Delivering unmatched performance, resilience and

security for your critical business workloads

IBM Confidential


Analytics & DB2

• New scalar functions and services and improved columnar function for EVI access

• Native/VIOS attach of FlashSystem 900

• Access Client Solutions – Performance Centre & SQL Enhancements

Extending IBM i for Mobile, Social and other Open Source Solutions

• Enabling Linux Little Endian on PowerVM for co-location of new Linux applications

• Enhanced application platform with Java 8, Python, Samba, node.js support

• i Access Mobile client for native Mobile access

• Enhanced IBM Collaboration solutions (e.g., Domino, Traveler, Sametime, Connections)

Availability & Virtualization• Backup Recovery Media Services(BRMS) modernizes automatic storage tiering

Hardware Enablement• Support S822 and PurePower

Selected Headliner 2015 TR enhancements for IBM i environments

Business Solutions Made Easy: IBM i 7.2


SAP strategy for HANA

From:• One DB per application• Point-to-point integration • Long running queries, e.g.

in batch mode

To:• One DB per landscape• No integration necessary• Real time execution


In Memory Computing: Re-think Paradigms

In-Memory Computing Imperative: Avoid movement of detailed data Calculate first, then move results

ApplicationLayer

DatabaseLayer

Calculation

Calculation

Today Future

14


SAP HANA Deployment Options for Different Use Cases

e.g. SAP ERP: CO-PA e.g. SAP BW -7.3 SP5-on HANA, Business Suite, S/4HANA

e.g. Demographics & Google Maps

e.g. SAP ERP: Operational reporting


17

• Modern ERP systems are challenged by mixed workloads, including OLAP--‐style queries. For

example:

• OLTP--‐style: create sales order, invoice, accounting documents, display customer master data

or sales order

• OLAP--‐style: dunning, available--‐to--‐promise, cross selling, operational reporting (list open

sales orders)

• But: Today’s data management systems are optimized either for daily transactional or analytical

workloads storing their data along rows or columns

• Drawbacks of the OLTP and OLAP separation:

• OLAP system does not have the latest data

• OLAP system does only have a predefined subset of the data

• Cost--‐‐‐‐intensive ETL process has to synch both systems

• There is a lot of redundancy,

• aggregates and materialized views required for performance

• Different data schemas introduce complexity for applications that combine sources

OLTP vs OLAP


SAP HANA

Column Store

Row Store

Analytics vs. Transactional Systems Differences

SAP BW

SAP HANA

Column Store

Row Store

SAP ECC

SAP HANA

Column Store

Row Store

SAP HANA

Column Store

Row Store

SAP HANA

Column Store

Row Store

A - F G - K L - Q R - Z Table Partitioning

Parallel Queries


SAP HANA on Power Architecture

2010 2011 2012 2013 2014 2015

Not a separate fork of the HANA code

HANA

Powerx86

OLAP Engine

Join Engine

Calculation Engine

HW Engine


SAP HANA on IBM POWER

+

Enterprise ready

Mission Critical Reliability

Flexiblility

Virtualized


SAP HANA on Power – Expected Customer Value

• Delivering the reliability required by mission critical systems

– HANA needs 30 to 40x memory of conventional DBs – Power delivers comprehensive memory protection by default , not as an option

– Historic fault resilience, dynamic component deallocation, on the fly repair and fault isolation are unmatched in open systems

• Performance

– Packing more HANA throughput into few cores leaves room for other requirements, reduces datacenter footprint and environmentals

– Over 4x memory bandwidth, 8 threads per core deliver consistent HANA response time, even during Delta Merge processing

• Flexibility

– Industry leading, low overhead virtualization included by default, not an option with high overhead and artificial constraints unsuitable for production workloads

– Capacity on Demand

– Partition isolation allowing prod, non-prod and HA to be intermixed


HANA on POWER Portfolio

S812L

S822L

Linux only Systems

S822

S824

Scale-Out

S824L

E870

Enterprise

E880

E850

HANA on Power

Solution Editions

S814

� IBM fast-start option

� Industry best-practice

� Quick configurations

� Aggressively priced

� Rapid deployment of SAP HANA on IBM Power

Systems


MemoryBuffer

DRAMChips

DDR Interfaces

POWER8

Link

Scheduler &

Management

16MB

Memory

Cache

“On-the-fly” lane isolation/repair

POWER8 Memory

10 chips per rank for extra redundancy beyond chipkill


RAS Feature POWER8 x86

Application/Partition RAS

Live Partition Mobility Yes Yes

Live Application Mobility Yes Yes, support issues

Partition Availability priority Yes No

System RAS

OS independent First Failure Data Capture Yes EX – MCA Recovery

Memory Keys (including OS exploitation) Yes No

Processor RAS

Processor Instruction Retry Yes No

Alternate Processor Recovery Yes No

Dynamic Processor Deallocation Yes No

Dynamic Processor Sparing Yes No

Memory RAS

Chipkill™ Yes Yes, some vendors

Survives Double Memory Failures Yes Yes, optional

Selective Memory Mirroring Yes No

Redundant Memory Yes Yes

I/O RAS

Extended Error Handling Yes No

I/O Adapter Isolation (PI-Bus and TCEs) Yes No

See the following URLs for addition details:http://www-03.ibm.com/systems/migratetoibm/systems/power/availability.htmlhttp://www-03.ibm.com/systems/migratetoibm/systems/power/virtualization.html

POWER8 RAS vs. x86


192750

172450

137010

126980

4818

2395 2283 2116

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

200000

Nav

igatin

ste

ps/H

ou

r/Co

re

Ad

-Ho

c N

av

igati

on

Ste

ps/H

ou

r

New World record set by IBM Power E870 on SAP BW Enhanced Mixed Load Standard Application Benchmark

with 2 Billion records

2Xper core

(1) IBM Power Enterprise System E870 on the SAP BW Extended mixed load standard application benchmark running SAP Netweaver 7.31 application; 4 processors / 40 cores / 320 threads, POWER8; 4.19GHz, 1024 GB memory, 192.750 adhoc navigation steps per hours on SuSE Linux Enterprise Server 11 and SAP Hana 1.0, Certification #: 2015024 Result valid as of June 1, 2015. Source: http://www.sap.com/benchmark.(2) Dell PowerEdge R930, on the SAP BW Extended mixed load standard application benchmark running SAP Netweaver 7.31 application; 4 processors / 72 cores / 144 threads, , Intel Xeon Processor E-7 8890 v3 ,2.5 GHz; 1536 GB memory, 172.450 adhoc navigation steps per hours on SuSE Linux Enterprise Server 11 and SAP Hana 1.0, Certification #: 2015014(3) Dell PowerEdge R920, on the SAP BW Extended mixed load standard application benchmark running SAP Netweaver 7.31 application; 4 processors / 60 cores / 120 threads, , Intel Xeon Processor E-7 4890 v2 ,2.8 GHz; 1024 GB memory, 137,010 adhoc navigation steps per hours on SuSE Linux Enterprise Server 11 and SAP Hana 1.0, Certification #: 2014044(4) HP DL580 Gen8, on the SAP BW Extended mixed load standard application benchmark running SAP Netweaver 7.30 application; 4 processors / 60 cores / 120 threads, , Intel Xeon Processor E-7 4880 v2 ,2.5 GHz; 1024 GB memory, 126,980 adhoc navigation steps per hours on SuSE Linux Enterprise Server 11 and SAP Hana 1.0, Certification #: 2014009

SAP and all SAP logos are trademarks or registered trademarks of SAP AG in Germany and in several other countries. All other product and service names mentioned are the trademarks of their respective companies.

IBM E870

POWER8 4p/40c/320t

DellPowerEdge

R930E7-8890 v34p/72c/144t

HPDL580 Gen8E7-4880 v24p/60c/120t

DellPowerEdge R920

E7-4890 v2,4p/60c/120t

12% faster

© 2015 International Business Machines Corporation 26

POWER8 processor is Purpose Built – resulting in premium performance over Intel Xeon

Haswell EP

E5-26xx v3

Ivy Bridge EX

E7-88xx v2

Haswell EX

E7-88xx v3POWER8

Clock rates 1.7-3.7GHz 1.9-3.4 GHz 2.0-3.0 GHz 3.0-4.15 GHz

SMT options 1, 2 1, 2 1, 2 1, 2, 4, 8

Cores per socket 18 15 18 12

Max Threads / socket 36 30 36 96

L1 Cache/core 32KB I + 32KB D3 32KB I + 32KB D3 32KB I + 32KB D3 32KB I + 64KB D

L2 Cache/core 256 KB 256 KB 256 KB 512 KB

L3 Cache/socket 45 MB 37.5 MB 45 MB 96 MB

Max L4 Cache/socket 0 0 0 128 MB

Performance Mode Memory Bandwidth

50-66.6 GB/s1 66.2-83.3 GB/s1 100 GB/s1 N/A

RAS Mode Memory Bandwidth

N/A 41.65-50GB/s1 41.652-58.3GB/s1 190-230 GB/sec

1) Depending on DIMM Rank and # of DIMMS/Channel

2) Lenovo supports a minimum of 50GB in RAS mode with Haswell-EX

3) L1 data cache with parity protection – 32KB, ECC – 16KB


TDI – Tailored Datacenter Integration

27

• Solution validation done by SAP and partner

• Preconfigured hardware set-up• Preinstalled software

• Installation needs to be done by customer

• Customer aligns with the hardware partner on individual support mode

Fast ImplementationSupport fully provided by SAP

More FlexibilitySave IT budget and existing investment

HANA Appliance


Delivering freedom of choice to the HANA Landscape

28

SAP HANA®

OSLinux Enterprise Server

Priority Support for SAP applications

File System Monitoring

Backup

Sys MgmtGPFS2, XFS

Server HardwarePOWER7+(non-prod) or POWER8

Standalone or shared/PVM virtualized

Applications

Storage Hardware

Customer choice

High AvailabilitySAP HANA Host Auto-failover & System

Replication, SUSE HA Ext, RH HA Plugin1, Symantec HA1, Tivoli SA

ECC

BW HA

PowerVM

ECC App

Hypothetical Power Landscape

CRM

ECC HA

BW

PowerVM

Data

ECC, BW, SM App Sbx

ECC QA

BW QA

CRM QA

ECC App

VIO

S

VIO

S

ECC HA QA

HANA

HANA

App

DB/App

DB/App

DB/App

App

HANA

HANA

App

HANA

HANA

App

HANABW HA QA

BW App

VIO

S

VIO

S

1) Not currently supported with Linux on Power. Contact RedHat for product plans.

2) GPFS is currently supported for Linux on Power by IBM. Contact SAP for plans for certification with HANA on Power..


SAP HANA HA: System Replication

Performance Optimized Cost Optimized

• Very low RTO• High cost for 100% standby

resources• Ideal for non-virtualized

systems and/or expectation of frequent outages

• With SPS10, secondary available Read Only

• Much lower cost – 10% reserved

• Works best with virtualized infrastructure

• If using VMware, may not be compatible with production (1 VM only, size restrictions)

• RTO from disk much slower• Memory must be reassigned

upon HA incident, very slow

Dedicated Shared

Performance and Cost

Optimized with POWER

• Very low RTO• Majority of secondary CPU

resources available to other LPARs

• Perfect fit for strong production level virtualization solution such as PowerVM

Shared

QA/DEVrunning


SAP HANA Delta-Merge Process

31

• Column store table - comprised of a Main storage and a Delta storage all located in RAM

• Delta storage optimized for write operations• Main storage is optimized for reads

• Use of Delta tables addresses performance issues of loading directly to compressed columns.

• Merge process takes the data out of the Delta structures and puts it into the Main structure

• This is a very CPU/Memory intensive task (!!!)Power delivers up to 5x

memory bandwidth, 4x CPU threads vs. x86!


When faster I/O may make a difference

Load time dependent on throughput of IOPS

Response time of on-demand columns dependent on latency, throughput of IOPS

Conventional DB

ClientPart number

Order date

Shipped date

Delivery date

QuantityItem Weight

Item cost

Package weight

Total price

Instructions Status

abc 123 8/1/2015 9/6/2015 9/15/2015 15 0.56 1.35 9.5 $42.00 Leave on porch In-transit

Delta Table in HANA

Single Async IOP

abc

123

8/1/2015

9/6/2015

9/15/2015

15

0.56

1.35

9.5

$42.00

Leave on porch

In-transit

Async IOPS = number of columns(at minimum)

Delta merge

To log file and in-memory row

abc 123 8/1/2015 9/6/2015 9/15/2015 15 0.56 1.35 9.5 $42.00 Leave on porch In-transit

RTO after failure or HANA patching


HWCCT – Hardware configuration check tool

Determine if system meets KPI requirements

• Landscape test• OS config validity• Consistency of

landscape based on reference architecture

• File system throughput/latency

• Network throughput for multinode configurations

• 9.5 GBits for single stream

• 9.0 GBits for duplex stream


Technical design and infrastructure deployment

� Planning and configuration

� Installation and verification

� Coordinate with SAP

� Validation and testing

� Recovery/availability plan

� Coaching and skills transfer

Operation and optimization

� Proactive services

� Solution support

� Optimization

� Service management

� Operating model

Assess, Upgrade and Migrate

� Project services

� Health check assessment

� Platform upgrade

� Deploy scale-out hardware1

Life cycle management

1 Scale our solution with General Parallel File System (GFPS)

IBM Services for SAP HANA on Power

34


Full SAP HANA on POWER Integrated Solution Service

Overview TSS HANA on POWER Total Solution Service

Individual

continuous customer interaction

Integrated HW and SW1

supportImprove availability & reduce resolution

time

Optimized access to a

comprehensive solution support

Avert problems:

proactive services, preventative support

advice & services

• This Service Component is a key differentiator for our POWER Platform• Close Linkage with SAP (Electronic Retain Ticket Interface with SAP )• With single-source and central responsibility, TSS services are designed to

simplify support and accelerate problem resolution for mission critical HANA on POWER environments.

• SUSE License can be included in the service package. This combination is a huge added value for our customers from an pricing and content perspective.

• For offering support and individual pricing contact your local TSS representative listed on next page.

SAP

SUSE


SAP HANA on IBM Power Summary

• Analytics enable decision making, Suite on HANA runs the business

– Power Systems delivers the non-stop operations required of these mission critical systems

• Performance - HANA demands extreme memory and thread throughput

– Power Systems delivers 4x threads, over 3x memory throughput compared to x86

• Flexibility– Power Virtualization built in, not a poorly supported afterthought

© 2015 International Business Machines Corporation 37

Merci

Grazie

Gracias

Obrigado

Danke

Japanese

French

Russian

German

Italian

Spanish

Brazilian PortugueseArabic

Traditional Chinese

Simplified Chinese

Thai

Korean

Hindi

DankieAfrikaans

SiyabongaZulu

Tamil

תודהHebrew

Blog: SAPonPower.wordpress.com


Open innovation to put data to

work in a waitless world

To find out more about how the IBM Power Systems platform can help meet your enterprise innovation needs, visit: www.ibm.com/power

Documents

POWER8. Open Innovation to Put Data to Work for SAP · PDF filePOWER8. Open Innovation to Put Data to Work for SAP ... SAP HANA Deployment Options for Different Use Cases ... Open