itg1505 power8 linux migration - digitaltrend technology group may 2015 strategic considerations for...

International Technology Group 9128 Strada Place, Suite 10115

Naples, Florida 34108-2931 Telephone: 888.513.0222

Email: [email protected] Website: ITGforInfo.com

Management Report May 2015

Strategic Considerations for Migrating x86 Linux Applications to IBM POWER8

International Technology Group May 2015


Table of Contents EXECUTIVE SUMMARY 1

Key Questions 1 Why POWER8? 1 Migration Options 3 Costs and Risks 4

BEST PRACTICES 5 Overview 5 General Principles 5 Project Phases 6 Tools and Services 8

PROCESSORS 11 Market Dynamics 11 Intel Evolution 11 POWER8 13

Power Systems 13 POWER7 and POWER8 14 RAS Features 14 POWER Futures 15 PowerVM Virtualization 15

KEY SOLUTIONS 17 Overview 17 Open Source Software 17 Analytics Appliances 18

APPENDIX: OPENPOWER FOUNDATION 20 Overview 20 Members and Activities 20

Platinum Members 20 Gold Members 21 Silver Members 22 Academic Members 25

List of Figures 1. IBM POWER8 and Intel Haswell Processors 2 2. Industry Experience – Reasons for Project Shortfalls 5 3. Representative Migration Project Phases 7 4. Tools and Services for x86 to POWER8 Linux Migration 10 5. Additional IBM Business Partner Services for x86 to Power Linux Migration 10 6. Per Core Performance for Intel and IBM POWER8 Processors – SAP SD Two-Tier Benchmarks 12 7. Recent Intel Processor Evolution 13 8. IBM POWER7+ and POWER8 Processors Summary 14 9. Unique POWER8 RAS Technologies 15

10. PowerVM Architecture 16



Executive Summary

Key Questions IBM’s POWER8 has generated headlines since its launch in August 2013. It was – and remains – the industry’s most powerful and sophisticated processor design. It also marked a major shift in IBM server strategy. The POWER8 processor and POWER8-based systems are designed to run Linux, KVM, OpenStack and other open source software.

IBM decided to open the POWER8 architecture. The OpenPOWER Foundation allows vendors and users full access to this architecture. The Foundation has grown to 100+ members, including leading hardware manufacturers and software developers worldwide. A five-year IBM investment of $1 billion is also helping to create an ecosystem of value-added enhancements, ISV and service provider support, and skill bases.

POWER8 has changed the competitive picture in microprocessors. For many users and vendors, POWER8 is the strongest challenger – and the main alternative – to Intel x86 architecture. ARM processors remain confined to low-end applications. POWER8 is designed for high-performance server-based computing.

Early experiences have shown that most open source software written for x86 Linux can be migrated – often with no more than a simple recompile – to Power Linux. Users have reported migration times ranging from a few hours for simple applications to five months for a full TurboLAMP stack containing 40,000+ binary packages.

For x86 Linux users and developers, two key questions are posed: (1) why migrate applications to POWER8, and (2) how to do so?

This document addresses these questions. It outlines the potential advantages of employing IBM POWER8- rather than x86-based servers, and presents best practices that can enable organizations to minimize costs, difficulties and risks of migration to Power Linux.

Why POWER8? Although IBM has supported Linux on Power Systems since 2001, the POWER8 introduction has seen technology changes that make it a great deal more attractive to x86 Linux developers. Unlike its predecessors, POWER8 supports little endian as well as big endian byte structures. As most x86 Linux applications support little endian, migration has become a great deal faster and easier.

POWER8-based systems are, moreover, differentiated from Intel-based equivalents in four main areas:

1. Performance. POWER8 offers baseline performance 2 to 2.5 times higher per core than current-generation Intel processors. The performance gap expands when co-processors are employed, and when software stacks are fully optimized. For analytics workloads, performance disparities of up to 82 times have been reported.

Costs for software charged on a per core basis may be significantly lower than for Intel equivalents. This picture is unlikely to change during the lifecycles of Intel Haswell, Broadwell and Skylake processors. For at least the next three years – and probably for a great deal longer – it can be expected that POWER8-based systems will continue to outperform Intel equivalents by wide margins.

While Intel processors are designed to support a wide range of server, PC and mobile device requirements, POWER8 is aimed squarely at workloads such as analytics, big data and real-time processing. Microprocessor and system-level designs are significantly different, and are expected to diverge further during POWER8+ and POWER9 lifecycles.


Strategic Considerations for Migrating x86 Linux Applications to IBM POWER8 2

POWER8 processors, for example, support larger numbers of threads per core (four times compared to Intel Haswell), main memory (also four times), memory bandwidth (2.3 times), clock speed and cache. In the POWER8 design, L1 through L3 cache are on-chip, while 128 gigabytes (GB) of L4 cache – there is no Intel equivalent – is added off-chip.

The overall POWER8 design is, as figure 1 illustrates, a great deal better optimized for workloads requiring fast movement of data between processors, memory and I/O. POWER8 is, as IBM claims, a data-intensive design.

INTEL HASWELL IBM POWER8

Introduced 04/14 04/14

Cores/chip 18 12

Max threads/core 2 8

Max clock rate 3.5 GHz 4.1 GHz (1.2x)

Max main memory 4 TB 16 TB (4x)

Max L-1 cache 64 KB 64 KB

Max L-2 cache 256 KB 512 KB (2x)

Max L-3 cache 45 MB 96 MB (2.1x)

Max L-4 cache N/A 128 MB

Max memory bandwidth 68 GB/s 410 GB/s (2.3x)

Max sockets 8 8 (16 planned)

Process technology 22nm 22nm

Figure 1: IBM POWER8 and Intel Haswell Processors

POWER8 also responds to a broader industry shift. Server performance has become increasingly defined not only by core CPUs, but also by co-processors. In this area, POWER8 again outclasses Intel designs.

2. CAPI interface. Use of accelerators based on field programmable gate arrays (FPGAs) and graphics processing units (GPUs) have been widely employed on x86 servers.

POWER8, however, offers a major differentiator – the Coherent Accelerator Processor Interface (CAPI) allows accelerators to be tightly integrated with core CPUs, and to operate at memory speeds. Data latencies are significantly reduced. The performance impact is often dramatic.

The CAPI interface is being exploited by members of the OpenPOWER Foundation. These include leading suppliers of FPGA- and GPU-based accelerators, interconnects and adapters, flash memory devices, servers and boards, and other products and services.

The OpenPOWER Foundation also includes open source software providers, and academic and public sector institutions engaged in research into supercomputing and open software development.

OpenPOWER Foundation activities are reinforced by IBM investments in development, migration and support for POWER8 and Power Linux. A wide range of tools and services, described later in this report, have been made available to users and developers.

3. System architecture. Although POWER architecture was originally developed to run the IBM version of UNIX, AIX, the company has abstracted key hardware- and microcode-based features so that they apply equally to Linux environments. As a result, POWER8-based systems offer capabilities previously found only on UNIX servers.

POWER8-based Linux systems are significantly differentiated from x86 equivalents in such areas as virtualization, multitasking and workload management.



One implication is that POWER8-based systems can support higher workload densities; i.e., more partitions, operating more efficiently, can be supported than on equivalent x86 servers. Industry experience has been that two to three times more virtual machines (VMs) can be supported on POWER8-based systems.

(This explains much of the interest shown in POWER8 by cloud and hosting providers. For this group, higher workload densities translate into major bottom-line savings. User organizations supporting large numbers of VMs may expect comparable gains.)

4. Reliability, availability and serviceability (RAS). POWER8-based systems benefit from hardware- and microcode-based RAS features derived from earlier Power Systems as well as IBM mainframes, which are generally recognized to offer the highest levels of availability of any major platform worldwide.

These are generally more integrated and effective than x86 equivalents, employ more advanced microelectronics technologies, and include unique capabilities not found in Intel-based designs.

POWER8 strengths are reflected not only in IBM servers, but also in the growing field of hardware platforms offered by OpenPOWER Foundation members. Systems that have been introduced or are under development range from single-processor units to mainframe-class systems and next-generation supercomputers with five to ten times the performance of today’s largest machines.

POWER8 has also been adopted by large cloud and hosting providers worldwide, including Google, Rackspace and OVH. For this group, POWER8 represents new business opportunities as well as a new means to build higher-performance, lower-cost server farms.

These options are, however, not restricted to Internet giants. They are available to any organization, of any size, in any industry.

Migration Options The amount of time and effort required to migrate x86 Linux applications to POWER8-based systems varies depending on size, complexity and technical content.

Where applications have no platform-specific dependencies, a simple recompile may be sufficient. Users, who have done this, report that migrations were completed in a few hours…in a morning…in a day.

At the other end of the spectrum, a large company expects to complete migration of 200-+ internally developed and ISV systems to Linux on Power in less than 18 months.

Although conditions for individual users may vary, experiences with POWER8 Linux migrations since 2013 suggest some general rules of thumb. These include the following:

• Most x86 Linux applications with no platform dependencies will run as is on POWER8, requiring only recompiles.

• Most (IBM says over 95 percent) of Linux applications for any platform written in C++ will not require source code changes; i.e., again requiring only recompiles.

• Applications that contain hardware-specific dependencies should be evaluated on a case-by-case basis to determine levels of cost and difficulty. Customized features, databases and middleware, use of CAPI-attached devices and other variables should also be taken into account.



For organizations seeking to migrate x86 Linux applications, IBM supports POWER8-optimized versions of Red Hat Enterprise Linux (RHEL) 7 and 12, SUSE Linux Enterprise Server (SLES) 12 and Ubuntu 14.04 and 14.10. For these distributions, POWER8 has release parity with x86; i.e., updates will be released at the same time.

Organizations should study applications carefully before starting migration projects. Proper allowance should also be made for planning, preparation, testing and tuning, go-live and post-migration activities.

As with any IT project, odds of success will increase if best practices – meaning documented practices derived from experience – are employed. A representative set of these is provided in the next section of this report.

Costs and Risks Despite some early concerns that IBM would not be able to match Intel pricing, this has so far not proved to be the case. IBM’s decision to open the architecture to third parties has fundamentally changed competitive economics. POWER8 has drawn support from many of the world’s largest high-volume, low-cost hardware manufacturers.

It can be expected that costs of POWER8-based systems will remain close to those of Intel-based. Equally, high levels of server performance – which are magnified by support from industry-leading co-processor suppliers – move price/performance competitiveness to a new level. This will clearly continue to be the case for the foreseeable future.

The breadth of third-party support that has been reflected in the OpenPOWER Foundation indicates that POWER8 momentum is building. Current design cycles suggest that dozens, potentially hundreds of new POWER8- and CAPI-based designs will appear during 2015 and 2016. Compatibility between POWER8 and x86 Linux, and aggressive recruitment of the open source community, mean that availability of software will not retard this momentum.

For organizations interested in knowing more about POWER8, costs and risks of entry are minimal. Development, test and proof of concept (POC) servers may be purchased for a few thousand dollars. Free or low-cost cloud development platforms are currently offered by IBM, OVH and (for Ubuntu) Site Ox.

The best of the best practices is simple: try it and see.



Best Practices

Overview A wide range of migration techniques and methodologies are available from vendors, ISVs, consultants and industry groups, and many user organizations have developed their own. The following are based on general industry experiences for migration projects, and on inputs from organizations that have migrated x86 Linux to POWER8-based systems.

Not all practices are applicable to all projects. It is, however, recommended that users should be aware of all factors potentially affecting cost, difficulty and risk exposures. Most project best practices have been documented for years, even decades. Yet project shortfalls continue to occur.

Figure 2, which is drawn from a recent survey of challenged corporate IT projects, illustrates some of the most common reasons for these.

Figure 2: Industry Experience – Reasons for Project Shortfalls

The principal causes of shortfalls such as delays, cost overruns, and inadequate functionality are not technological. They have to do with people, processes and organizational buy-in. If these are not adequately dealt with, projects may damage the credibility of both the solution, and of those who promoted and supported it. Equally, if they are addressed, the reverse will typically occur.

General Principles For organizations migrating x86 Linux applications to POWER8-based systems for the first time, some general principles apply.

Odds of success will be highest when:

1. Organizations have the required skills and experience for migration internally, or draw upon outside assistance to provide these.

2. Applications are simple rather than complex.

3. Applications are migrated as is, without functional changes.

4. Applications are not business-critical and/or sensitive to downtime.

39%

27%

33%

37%

47%

48%

54%

63%

66%

74%

Other

Consultant problems

Staff turnover

Vendor/product problems

User resistance/adaptation problems

Unstable/immature technology

Inadequate project management

Poor initial requirements definition

Underestimated costs &/or difficulties

Requirements/priorities changed during project

Base: 230 corporate projects Source: International Technology Group



5. Data migration challenges are simple. For a first project, organizations should focus on like-to-like migrations (e.g., NoSQL to NoSQL, relational to relational), of the same database management system.

6. Projects are properly structured, managed and documented. Selection of an application or applications to migrate, along with planning, preparation, testing, tuning, go-live and post-migration activities all require attention. If sufficient attention is given to these, risk exposure will be significantly lower.

7. Project durations are short. The longer a project lasts, the greater the risk that schedules will slip, costs will overrun, scope creep will occur as user requirements change, key personnel will not be available and other disruptions will occur.

8. Management supports the project, and is prepared to commit sufficient resources to complete it. For longer projects, it is important to maintain management support over time. Personnel and/or priorities may change, and attention spans are sometimes short.

9. End users are engaged in the project – not only during requirements definition, but also during testing, tuning, go-live and post-deployment activities.

As a further general principle, projects will be simpler and less risky if co-processors such as GPUs and FPGAs are not employed. However, applications employing these will tend to show the largest performance gains, at least for high-end workloads. Organizations should understand the trade-offs involved, and ensure that appropriate skills and experience are available.

Project Phases A representative list of project phases is shown in figure 3. Again, not all stages and activities may be applicable to individual projects, but organizations should be sensitive to the issues involved.



1. Research Ø Check with vendors, ISVs, service firms & other users. Ø Check references – try to speak to users directly, & verify that their experience is applicable to your own situation (e.g., application

characteristics, project cost & duration, project methodology & resources, anything they would have done differently?). Ø Check potential sources of outside assistance; e.g., OpenPOWER Foundation members.

2. Basic choices Ø Employ in-house staff or outside service provider?

o Key question: does your organization have the skills, experience & resources to perform application & data migration in-house, or should outside assistance be brought in?

o If external services are used, in-house staff should be involved & provision should be made for skills transfer. Ø Employ on-premises systems or development cloud? (Or both, for different project phases). Ø Migrate custom, ISV or hybrid applications?

3. Application Selection Ø General factors to consider:

o Cost/difficulty of migration. - Personnel & system resources required – & management willingness to commit these.

o Project duration (may be approximate at this point). o Business-criticality/sensitivity to downtime. o Management/user support for migration. o Risk of impaired solution, schedule & cost overruns.

Ø If choosing between multiple applications, rate difficulty, cost & risk based on comparative scale (e.g., 1-10). Ø Be prepared to revise selections if risk, cost or other factors are not favorable.

4. Application Evaluation Ø Basics:

o Examine code base & documentation. o Consult with original developer(s) & deployment team. o Determine hardware dependencies. o Recompile vs. broader changes? o Size, structure & features (simple/complex?)

- Database/file system? o Language/development tools used. o Customization? o Third-party software components; e.g., middleware. o Software versions (updates required?)

- Functional upgrade(s) required? o Backup & recovery arrangements. o High availability/failover requirements.

Ø Employ tool(s) that generate reports on which application features may be upgraded automatically, & which will require manual work. Ø Consult end users – understand their expectations for the migrated application.

5. Project Planning Ø Basics:

o Determine skills required – training & experience. o If appropriate, determine nature & timing of external assistance required. o Designate project team & project manager. o Determine project duration. o Set phases, checkpoints & milestones – ensure that these are clear & verifiable.

- (If third parties are supplying services, ensure that payment is tied to these milestones.) o Ensure key staff availability during project. o Ensure hardware & software availability (e.g., in-house, virtual loaner) during project. o Identify additional software requirements; e.g., new or upgraded systems software licenses, migration & tuning tools. o Plan for other requirements (e.g., development cloud reservations).

Ø Develop plans for all project phases; e.g., preparation, migration, testing, deployment, go-live & post-migration activities. Ø Consider potential contingencies (e.g., technical problems, schedule & cost overruns) & how to would deal with them. Ø Estimate cost (in-house & third-party personnel, hardware, services). Ø Prepare & agree upon written project plan, including tasks, schedules, milestones, staff allocation & configuration details. Update for

later changes. Determine responsibilities for tasks.

Figure 3: Representative Migration Project Phases



6. Preparation Ø Form project team with required skill sets. Ø For outside assistance – select & contract with supplier(s), specifying milestones & availability of personnel & resources. Request &

inspect contractor resumes. Ø Ensure staff is familiar with platform, software stack & tools – required training on these should be conducted before migration starts. Ø Ensure software versions are up to date – version upgrades should be conducted before migration starts. Ø Attempt to freeze functional & technical specifications during project. If this is not possible, minimize & control changes. Ensure

management & users understand & accept risk implications. Ø Recheck project plans, schedules, milestones, & personnel & system resource commitments, & ensure that these are agreed before

migration begins. Ø Ensure all code & documentation is up to date & available to project team. Ø Implement backup & recovery plans. Ø Set up user mail list. Send notifications of downtime & go-live schedule in advance. Also send later reminders of key dates & times. Ø Make sure at least two people are familiar with migrated application, software stack & underlying platform in case of turnover.

7. Migration Ø Create & update documentation on ongoing basis – ensure all procedures can be repeated. Ø Avoid changes to project scope if possible. If changes cannot be avoided, use change management tools/processes. Ø Ensure ongoing backup & recoverability of changed software in case of errors or data loss. Ø Automate tasks where feasible.

8. Testing & Tuning Ø Test solution & tune accordingly. Expect multiple cycles until performance & functionality goals are met. Ø Depending on applications, plan for functional, upgrade, integration, performance, volume & load, &/or other tests. Ø General principle: Test! Retest! Test again! Ø Involve users directly in testing to ensure all functional requirements are met. Modify & tune as appropriate.

9. Go-live Ø Develop detailed checklists of go-live processes & potential issues. Ø Set up intranet site & provide hot line for users to report problems. Respond accordingly. Ensure availability of key staff until problem

reports drop to normal level based on past experience. Plan for minimum of 48 hours – longer is better. 10. Post-migration Activities

Ø Monitor performance & quality of service (e.g., response time, availability). Analyze results & correct shortfalls. Ø Ensure contact is maintained with users through e-mail, intranet site &/or telephone. If the project is properly managed, it may serve to

develop closer & more productive relationships with end users. Ø Implement standardized change request & response procedures. Document & analyze user experiences. Ø Consult management as to whether application is meeting business requirements. Ø Institute regular newsletter to encourage interest among other managers & user groups in similar migrations.

All stages: Look for Warning Signs Ø Management &/or end-user interest wanes. Ø Pressure to meet deadlines causes organizations to accelerate project activities without adequate planning, preparation or testing. Ø Schedule, cost slippages occur at an early stage of projects, & deteriorate during subsequent phases. Ø Team morale declines, key personnel move on (to avoid being associated with a failed project?), people stop turning up to meetings,

e-mails &/or phone calls are not returned…

Figure 3 (cont): Representative Migration Project Phases

Tools and Services A wide range of tools and services from IBM as well as third parties are available to migrate Linux applications from x86 to POWER8. Figure 4 shows examples.



IBM TOOLS

IBM Software Development Kit for Linux on Power (SDK) Free Eclipse-based Integrated Development Environment (IDE) for development, build and debugging of big & little endian Linux applications on POWER8 systems. Integrates C/C++ source development with key IBM tools including Advance Toolchain for Power Linux, Post-Link Optimization (FDPR-Pro) and IBM Software Development Toolkit, Java Technology Edition.

Also bundles plug-ins from the Eclipse Linux Tools Project such as Autotools, Callgraph, ChangeLog, GCov, GProf, Libhover, Man Page, Oprofile, Perf, RPM Stubby, Specfile Editor, System Tap Editor and Valgrind.

IBM Installation Toolkit for Linux on Power Simplified, experience-based setup & workload tuning toolset for RHEL & SLES installation on Power. Includes 20+ value-added Linux on Power productivity & reliability, availability & serviceability (RAS) tools. Allows browsing & search of 60+ Linux guides & manuals, plus Power documentation. Works with new & existing configurations, allowing reuse of disks & partitions.

Advance Toolchain for Power Linux Open source (GNU) development & runtime toolset optimized for POWER8 systems. Enables key POWER8 features such as optimized scheduler, transactional memory, cryptographic operations & support for big endian (ppc64) & little endian (ppc64le). Includes GNU Compiler Collection (gcc, g++ & gfortran), C library (glibc), Binary Utilities (binutils), Decimal Floating Point Library (libdfp), AUXV Library (libauxv), Debugger (gdb), performance analysis tools (oprofile, valgrind, itrace), multi-core exploitation libraries (Intel TBB, Userspace RCU, SPHDE), support libraries (libhugetlbfs, Boost, zlib, etc.) plus RHEL-, SLES-, Ubuntu- & Fedora-specific features.

XL C/C++ & Fortran Compilers New (released 12/14) PGI optimizing XL C/C++ & Fortran compilers support big & little endian Linux distributions on POWER8 processors & architecture. Compilers provide a user interface, & language, parallel programming & optimization features identical to those available on PGI Linux x86 compilers. XL C/C++ compiler offers source compatibility with gcc while providing binary coexistence. Supports latest C11 & C++11 language standards. For little endian architecture, combines LLVM Clang front-end infrastructure with IBM optimization technology in back-end. New compilers support high performance computing (HPC) systems based on the IBM POWER architecture, including NVIDIA GPU-accelerated IBM POWER8 systems, & additional systems under development by members of the OpenPOWER Foundation.

FDPR-Pro IBM Research-developed tool providing additional workload-specific performance optimization after applications have been compiled & linked. Collects information on program behavior running customer workload, re-analyzes the program, applies global optimizations (including program restructuring), as well as more localized optimizations (e.g. function inlining) & creates a new version of the program based on its runtime profile. The modified program typically runs faster & uses less real memory.

IBM SERVICES Power Development Cloud

Offers no-charge remote access for up to 14 days to IBM hardware, including IBM POWER8 systems with Linux, for application development, porting & functional testing. Systems include virtualized CPU, disk & networking. IBM middleware may also be accessed.

Power Systems Linux Centers Centers provide a global network of resources supporting development & deployment of Red Hat, SUSE & Ubuntu on Power applications. Services include customer briefings; training workshops, seminars & courses; & proof-of-concept, prototyping, validation & integration testing, migration, & performance & scalability optimization on POWER8 systems. Services are available to users, ISVs & academic institutions, & include hands-on assistance in exploiting POWER8 parallel processing & advanced virtualization. Business Partner go-to-market services are also provided. Centers are located in Beijing, China; Austin, Texas & New York, New York USA; Montpellier, France; & Tokyo, Japan.

IBM Systems Lab Services 50+ experienced Linux on Power technical consultants team with ISVs & Business Partners to provide on-site assistance with technically complex client site situations representing qualified sales opportunities. Services, which are available worldwide, include proof of concepts, performance optimization, virtualization optimization & management & ongoing support.

IBM Migration Factory Custom service offered by IBM since 2003 migrates customer & ISV applications from competitive to IBM platforms, including POWER8-systems. Migration projects may be IBM-managed, or conducted by customers or third-party integrators with IBM assistance. Services draw upon dedicated technical specialists, as well as a database of customer experiences drawn from 8,000+ migration projects conducted for organizations of all sizes, in a wide range of industries. Migration Factory employs a five-phase process addressing assessment & planning, migration testing, tuning, & training & deployment.

Figure 4: Tools and Services for x86 to POWER8 Linux Migration



THIRD-PARTY SERVICES

Site Ox Cloud-based development platform for Ubuntu on POWER8. No-charge access is available for up to two weeks. Deployment is automated, & typically takes a few minutes. Site Ox is the official implementation partner of the IBM/Canonical collaboration.

RunAbove Public Beta for POWER8 Cloud-based POWER8 development systems offered by Web hosting firm OVH. US access is initially available for $32/month for 8 threads, 4 GB RAM & 10 GB local RAID; & $700/month for 176 threads, 48 GB RAM, 480 GB local RAID & VM support. Current pricing & availability should be verified.

Figure 4 (cont): Tools and Services for x86 to POWER8 Linux Migration

Additional services, summarized in figure 5, are available to IBM Business Partners.

IBM BUSINESS PARTNER PROGRAMS

IBM Innovation Centers 50+ Centers worldwide provide technical expertise for Linux on Power development, deployment & optimization. Centers offer classroom & online training, consulting, proof of concept/prototyping, testing, migration, & other technical services. Go-to-market assistance is also provided for Business Partners.

Rapid Port Program Offers in-depth technical support for application porting, along with IBM Power Systems & middleware for qualified sales opportunities. Services are typically at no cost.

Hardware Mall Allows discounted lease or purchase of IBM equipment to support development, testing, customer demonstration & related activities. Program is available in Europe & North America.

Figure 5: Additional IBM Business Partner Services for x86 to Power Linux Migration



Processors

Market Dynamics Intel dominates the microprocessor market. Advanced Micro Devices (AMD), once Intel’s main challenger in x86 processors, has been reduced to a niche player. Demand for UNIX servers based on IBM POWER and Oracle SPARC has also eroded.

Until recently, the principal emerging competition to Intel appeared to ARM processors, which have proved popular in smartphones and other mobile devices. Performance limitations have, however, restricted these to the low end of the microprocessor market.

As Intel’s market position has expanded, it has responded to a more diverse spectrum of customer requirements. Its principal microprocessor lines have been subdivided into low-end, midrange and high-end model groups built around common architectures. The company has also added its own GPUs and supercomputing features – most recently, through Advanced Vector Extensions (AVX), introduced with the Sandy Bridge design in 2011.

Intel’s success has, however, brought vulnerabilities. By stretching single basic designs to meet a growing range of demands, it has opened the door for specialists. Increasing use of third-party FGPA- and GPU-based accelerators has also diluted the company’s control over the x86 market.

The IBM POWER8 strategy has sought to exploit these vulnerabilities. The POWER8 design is aimed at server applications, in which it offers superior performance, workload density and availability. IBM has also, through individual vendor agreements and the OpenPOWER Foundation, been able to mobilize a great deal of support from established, as well as emerging accelerator vendors.

Large segments of the IT industry clearly want an alternative to Intel. This group ranges from large Internet and cloud service providers to ISVs, systems integrators and end users who must manage large x86 server populations. There is widespread concern that being locked in to Intel designs will impair future business growth and profitability.

More than any non-Intel platform introduced in the last decade, POWER8 has succeeded in leveraging these concerns.

Intel Evolution Over the last few decades, the IT world has witnessed extraordinary growth in the performance of x86 microprocessors. But conventional designs are starting to show diminishing returns.

Intel has remained firmly committed to Moore’s Law, which was first articulated by company co-founder Gordon Moore in 1985. According to Moore’s Law, the number of transistors in a dense chip doubles approximately every two years.

In practice, this has meant that Intel has focused on increasing chip density, placing more cores on a single silicon wafer over time. Process technology has evolved rapidly. The company began producing chips with a 22-nanometer process in 2011, will move to 14 nanometers in 2015, and expects to reach 10 nanometers in early 2017.

This strategy has, however, had a striking side effect. While the number of cores per chip has increased – the latest Intel Ivy Bridge and Haswell microprocessors employ up to 15 and 18 cores per chip respectively – per core performance has largely stagnated. This effect is illustrated in figure 6.



5,451

2,142

2,057

2,057

2,067

2,158

2,230

2,186

2,377

2,434

2,436

2,503

IBM Power E870, 8/80/640

Dell PowerEdge R720, 2/24/48


Cisco UCS B420 M3, 4/48/96

HP ProLiant BL660c, 4/48/96


Cisco UCS 460 M4, 4/60/120


HP ProLiant DL380 Gen9, 2/36/72

HP ProLiant DL380 Gen9, 2/36/72

Cisco UCS C240 M4, 2/36/72

PowerEdge R730, 2/36/72

Figure 6: Per Core Performance for Intel and IBM POWER8 Processors – SAP SD Two-Tier Benchmarks

Other industry performance metrics show similar pictures.

One implication of this trend is that, in many organizations, software costs have risen. Major database and middleware vendors typically price on a per core basis. Performance demands have caused numbers of cores to increase.

In contrast, IBM has placed a major focus on delivering higher performance per core. According to SAP benchmarks, POWER8 performance per core ranges from 2.2 to 2.5 times higher than Haswell, and 2.4 to 2.7 times higher than Ivy Bridge equivalents.

Intel product strategy shows no signs of changing. The company’s commitment to Moore’s Law as its main engine of performance improvement, and its tick-tock product cycle, allow for a certain predictability.

Tick tock – Intel’s term – means that a new microprocessor (tock) is followed by a process shrink; i.e., the same design is manufactured using a smaller die size (tick).

Ivy Bridge, for example, is a process shrink of Sandy Bridge, while Haswell is a new design, and Broadwell, low-end models of which have begun to appear in 2015, is a process shrink of Haswell. The next new Intel design, Skylake, may appear in 2H 2015, but will be in full production only during 2016.

Figure 7 illustrates this evolution.

In this presentation, Broadwell variables shown are for entry-level U series processors, the only models introduced to date. Broadwell availability appears to have been affect by production delays, and may be rapidly superseded by Skylake.

IBM is known to be planning upgrades of the POWER8 processor over the next few years. There is no reason to expect that the gap in performance per core will decrease.

Haswell EP 2.3 GHz

Ivy Bridge 2.4 to 2.8 GHz

POWER8 4.19 GHz

Processors/cores/threads SAPS/core



Ivy Bridge Haswell Broadwell (-U) Skylake

Processor type Process shrink Tick

New Tock

Process Shrink Tick

New Tock

Introduced 3Q11 3Q14 1Q15 2H15

Max cores/chip 12 (EP) 15 (EX)

18 (EP) 18 (EX)

2 (U) (?)

Max threads/core 2 2 2 2

Max clock rate 3.7 GHz (EP) 3.4 GHz (EX)

3.7 GHz (EP) 3.2 GHz (EX)

2.2 GHz (U) (?)

Max L-1 cache 32 KB 16KB w/ECC

64KB EX (?) (?)

Max L-2 cache 256 KB 256 KB EX (?) (?)

Max L-3 cache 30 MB (EP) 37.5 MB (EX)

45 MB (EP) 45 MB (EX)

EX (?) (?)

Process technology 22nm 22nm 14nm 14nm 10nm (2017)

Figure 7: Recent Intel Processor Evolution

POWER8

Power Systems POWER8 is the eighth generation of a reduced instruction set computing (RISC) processor architecture originally introduced by IBM in 1990.

During the later 2000s and early 2010s, POWER-based servers progressively gained share in high-end UNIX servers, accounting for more than 50 percent of this market by 2013. A number of trends, however, including mounting x86 Linux competition and weaknesses in growth markets, caused revenues to begin a sustained decline in 2012.

For more than a decade, IBM POWER designs have been characterized by CPU performance leadership, high-density virtualization and highly effective workload management. Power systems have also maintained a clear lead in availability. Industry surveys have consistently shown that downtime for unplanned as well as planned outages is lower than for x86 servers.

Older Power Systems typically run the UNIX-based AIX operating system. IBM i, a midrange operating system derived from the AS/400 platform, also continues to be used by around 150,000 customers, primarily SMBs. IBM plans to continue supporting both.

The company has, however, progressively moved toward Linux on Power Systems. Linux support on Power was first introduced in 2001. By April 2013, a year before the POWER8 introduction, Linux applications from more than 1,000 suppliers worldwide had been ported to or developed on this platform. The IBM move to Linux did not begin in 2014.



POWER7 and POWER8 POWER8 builds upon the POWER7 architecture, introduced in 2010. A half-generation upgrade, POWER7+, was added in October 2012.

Capabilities carried over to POWER8 included symmetric multithreading, on-chip cache (POWER7 was the first major commercial processor to employ this technology), compression of data in memory (memory capacity may be increased by 75 percent with less than one percent processor overhead) and transactional memory (a mainframe-derived feature accelerating performance for high-volume transactional workloads).

POWER8, however, differs from POWER7 in important respects. Apart from higher baseline performance and CAPI, POWER8 is designed to more effectively support multithreading, larger cache sizes and higher memory and I/O bandwidth. Differences are illustrated in figure 8.

POWER7+ POWER8

Introduced 10/12 04/14

Cores/chip 8 12 (1.5x)

Max threads/core 4 8 (2x)

Max clock rate 4.4 GHz 4.1 GHz



Max L-3 cache 80 MB 96 MB

Max L-4 cache – 128 MB

Max memory bandwidth 180 GB/s 410 GB/s (2.3x)

Process technology 32nm 22nm

Figure 8: IBM POWER7+ and POWER8 Processors Summary

Numbers of threads and cache capacity can be adjusted for individual workloads. Optimization may be handled automatically by the system, or by system administrators.

POWER8 embedded DRAM (eDRAM) technology allows higher densities in L3 and L4 cache (the norm is around three times) than standard static RAMs (SRAMs). System latencies are reduced.

In terms of baseline performance (i.e., performance realized without use of co-processors), benchmarks and user experiences indicate a 1.5 to 1.7 times increase per core compared to POWER 7+, and 2 to 2.5 times compared to POWER 7.

In practice, however, variables such as memory and I/O throughput also affect performance for data-intensive workloads. Most industry benchmarks do not adequately allow for these, although they will become increasingly significant as data sizes expand and real-time data management and movement become more pervasive.

CAPI is a superset of PCI Express 3.0. It GPUs, FPGAs flash memory, and network devices to share address spaces and memory with POWER8 processors as well as co-processors, and interface with these at memory speeds. CAPI, which plays a central role in the OpenPOWER Foundation, replaces the earlier IBM GX bus series.

RAS Features Although POWER8-based as well as x86 servers incorporate numerous RAS features, POWER8-based systems implement distinctive technologies that are not included in Intel designs. Examples are shown in figure 9.

A further capability, Live Partition Mobility, allows movement of active LPARs between systems without disrupting operations. Service interruptions of one or two seconds may occur due to network latency, but are rarely noticeable to users.



TECHNOLOGY DESCRIPTION

Chipkill Technology capable of detecting & correcting single-bit as well as 2-, 3- & 4-bit errors in memory devices, including cache & memory interfaces. Employs RAID-like striping of data across memory to provide redundancy & enable reinstatement of original data in the event of a failure. Significantly more reliable than conventional error correction code (ECC) technology. Can survive double memory failures in 8-DRAM groups.

First Failure Data Capture (FFDC) Employs 1,000+ embedded sensors that identify errors in any system component. Root causes of errors are determined without the need to recreate problems or run tracing or diagnostics programs.

Dynamic Processor Deallocation Detects error patterns indicating potential processor failures, & stops & deallocates untrustworthy processors.

Processor Instruction Retry Alternate Processor Recovery Partition Core Contained Checkstop

If an instruction fails to execute, the system automatically retries the operation. If the failure persists, the operation is repeated on a different processor. If this does not succeed, the failed processor is taken out of service (checkstopped). Only logical partitions (LPARs) supported by the failed processor are affected.

Dynamic Processor Sparing Allows alternates to be automatically activated as replacements for failed processors.

Partition Availability Priority In the event of a processor failure, reallocates LPAR-based workloads based on assigned priorities.

Figure 9: Unique POWER8 RAS Technologies

POWER Futures Further enhancements to the POWER8 generation of systems are expected to include higher clock-speeds and larger models – according to IBM the POWER8 symmetric multiprocessing (SMP) design point is for 16 sockets.

Although few details have been released on these, POWER8+ and POWER9 successors are planned. Power9 processors will be employed in a major supercomputer project for the U.S. Department of Energy (DOE) scheduled for completion in 2017-18.

IBM has also indicated that it will incorporate NVLink, developed by NVIDIA, in a future POWER processor design. NVLink is a higher-speed alternative to PCIe. According to NVIDIA, it will operate at 80 GB/s, five times faster than the current PCIe Gen3 x16 standard.

PowerVM Virtualization In addition to PowerKVM, IBM offers its PowerVM hypervisor for use with POWER8-based Linux systems. PowerVM, which was introduced in 2001, has been progressively enhanced to become one of the industry’s most efficient and robust virtualization environments. It is widely employed by Power AIX users to support very large, business-critical virtualized installations.

PowerVM supports three types of partitioning:

1. Logical partitions are microcode-based partitions that isolate workloads more effectively than software-based techniques. System resources used by LPARs may be dedicated, or shared according to application priorities. LPARs also provide additional security functions. This approach (a.k.a. hard partitioning) is not supported on x86 servers.

2. Micro-partitions are software-based partitions. They are typically employed to support instances requiring limited system resources, and to improve load balancing for large, complex workloads.

LPARs and micro-partitions are supported by mechanisms that allow processor, memory and I/O resources to be pooled and re-allocated in an extremely granular manner.

Business-critical workloads may run in dedicated LPARs, using dedicated physical processors. Other workloads may be executed based on assigned priorities using combinations of threads, partitions and shared processor pools.



3. Virtual I/O Servers allow operating system instances running in multiple LPARs to share a common pool of LAN adapters as well as FC, SCSI and RAID devices; i.e., it is not necessary to dedicate adapters to individual partitions. Virtual I/O Servers may be duplicated for redundancy.

The overall PowerVM architecture, which is illustrated in figure 10, allows users to manipulate a wider range of variables – including threads, processors, cache, main memory and I/O, multiple types of partition, multiple threads and dedicated or pooled processors – with higher levels of granularity and flexibility than any x86 hypervisor.

Figure 10: PowerVM Architecture

The implications are important. Partitioning creates the potential for high levels of capacity utilization. However, the extent to which this occurs in practice depends on mechanisms that allocate system resources between, and monitor and control workload execution processes across partitions. The more effective these mechanisms are, the higher the level of sustained capacity utilization will be.

Like PowerKVM, PowerVM executes directly on POWER8-based hardware.

PowerVM

RESOURCE SHARING Processors • Cache • Memory • I/O • Threads

Virtual tape Virtual disks

HYPERVISOR

LPAR

LPAR

LPAR

LPAR

Virtual LAN

VIRTUAL I/O SERVER VIRTUAL I/O SERVER

Physical processors

SHARED PROCESSOR POOL

Virtual processors

DEDICATED PROCESSORS

Physical processors

SHARED PROCESSOR POOL

Virtual processors

LPAR

Micro-partitions LPAR

Micro-partitions



Key Solutions

Overview As a general principle, any software written for x86 Linux can be migrated to POWER8, and large numbers of ports are under way by vendors and users. In some cases, IBM and/or other OpenPOWER Foundation members have cooperated to optimize key open source software.

Examples of such software include Linux distributions, an Ubuntu-based TurboLAMP stack, the Redis NoSQL database and Docker.

IBM has also cooperated with partners to deliver a number of POWER8- and CAPI-based analytics appliances. These solutions are described in this section.

Open Source Software Key solutions include the following:

• Linux distributions. In addition to RHEL and SLES, IBM has also supported Ubuntu 14.04, and cooperated closely with developer Canonical Inc. in multiple projects.

The attention paid to Ubuntu, according to IBM, reflects its popularity in cloud deployments. The company claims that Ubuntu’s scalability, as well as close affinity with KVM and OpenStack, has led to its deployment for many of the industry’s largest clouds.

According to the 2014 OpenStack User Survey, Ubuntu was the most popular operating system among this group, with a 64 percent share of production systems. Canonical also claims a more than 50 percent share of operating systems on Amazon Cloud.

A Debian GNU/Linux port supporting POWER8 little endian also recently became available, and migration of other distributions is expected.

• Ubuntu TurboLAMP stack is an open source, POWER8-optimized software solution developed by Canonical with MariaDB, Zend, Mellanox Technologies and IBM.

(LAMP refers to Linux, Apache, MySQL and PHP, a popular combination of open source technologies that has been widely deployed for Web services. LAMP and more recently TurboLAMP stacks have evolved a great deal. The Canonical offering is designed to support latest-generation Web as well as mobile applications.)

MariaDB acts as a compatible replacement for the MySQL database, whose evolution under Oracle management (it was acquired by the company in 2010) has generated controversy in the open source community. Mellanox Technologies supplies CAPI-attached high-speed interconnect technology.

The overall stack also includes Apache Web Server and popular open source applications such as the Apigility Zend framework; Magento and X-Cart for e-commerce; Drupat, Wordpress, Joomla! for content management; and SugarCRM. Ubuntu Juju service orchestration tools enable deployment.

• Redis optimization for POWER8 has been a longstanding OpenPOWER Foundation development focus. IBM has worked closely with Redis Labs, the leading commercial supplier of software and services for the in-memory NoSQL database Redis, which has proved popular with open source developers.



Redis Labs offerings are available through the IBM SoftLayer cloud. Redis is also the core database of the IBM Data Engine for NoSQL, which is described below, and has been adopted by other OpenPOWER Foundation members.

• Docker and IBM cooperate through a strategic partnership that includes joint development activities as well as availability of Docker Hub Enterprise– a repository of 60,000+ dockerized services – through the IBM SoftLayer cloud. IBM also plans to implement Docker across its Power and mainframe systems.

Other members of the OpenPOWER Foundation have also adopted or expressed interest in Docker for high performance computing (HPC) as well as commercial applications.

Docker allows use of software-based containers to develop, test and deploy applications as an alternative or complement to VMs. The name refers to both the product set and the open source start-up company that commercializes it.

Key Docker attractions include the ability to (1) reduce system overhead – containers do not require full operating system copies, and normally operate with stripped down minimal versions of these and (2) allow portability between different operating systems and system environments. There has been much interest in using Docker to move applications and workloads transparently between clouds.

IBM is also working to optimize a wide range of other solutions for POWER8 and CAPI. These include popular HPC applications such as SOAP3, NAMD, GROMACS, FFTW library and Quantum Espresso.

Analytics Appliances Key solutions include the following:

• IBM Solution for BLU Acceleration is based upon IBM’s DB2 10.5 with BLU Acceleration, which employs a synthesis of latest-generation analytics technologies.

These include columnar database structures, in-memory processing, data skipping (i.e., systems may avoid processing data unnecessary to specific queries), parallel and vector processing, memory and disk caching, and an Actionable Compression algorithm that allows data to be analyzed while it is being compressed.

Also, BLU Acceleration employs a dramatically simplified SQL design. The system does not, for example, employ schemas, indexes, aggregate tables or partitioning. It is built into the DB2 10.5 kernel using lightweight software structures that minimize overhead.

BLU Acceleration users have reported major gains in query performance – in some cases, by orders of magnitude – compared to conventional relational databases. The simplified SQL design has also meant easier application development, faster deployment and reduced DBA staffing.

Solution exploits key features of POWER architecture, including larger numbers of registers (POWER8 supports 64, compared to 16 for Intel-based designs) and Single Instruction/Multiple Data (SIMD) optimization.

• IBM Solution for Hadoop is built around the IBM Hadoop distribution, BigInsights. In addition to a full Apache Hadoop stack, this includes enhanced tooling for development, management, monitoring and administration, high availability, security and other functions. BigInsights is generally regarded as the most scalable and robust Hadoop distribution for large-scale enterprise deployments.



BigInsights also includes BigSQL, a native SQL query engine that allows developers to leverage existing SQL skills tools to query Hive, HBase or distributed file system data. Developers may use standard SQL syntax and, in some cases, IBM-supplied SQL extensions optimized for use with Hadoop.

Infrastructure enhancements include large-scale indexing (BigIndex), job scheduling (BigInsights Scheduler), workload management (Adaptive MapReduce) and grid computing (Platform Symphony).

BigInsights also offers an optional alternative to HDFS, IBM General Parallel File System – File Placement Optimizer (GPFS-FPO). This is a Hadoop-optimized implementation of the IBM GPFS distributed file system, which has been widely deployed for more than a decade for scientific and technical computing, as well as commercial applications.

• IBM Data Engine for Analytics includes IBM Cognos Business Intelligence (BI); SPSS modeling, predictive analysis and decision support tools; DB2 BLU Acceleration; and IBM InfoSphere Information Server (IIS) DataStage handling data extract, transformation and load (ETL). Exploits CAPI-attached Altera FPGAs.

• IBM Data Engine for NoSQL enables deployment of NoSQL data stores at a significantly lower cost than if x86 servers are employed. The solution – co-developed by Altera, Canonical, IBM and Redis Labs – combines BigRedis (a POWER8-optimized version of the Redis key value store), Ubuntu, Altera FPGA-based accelerators, and IBM POWER8-based and FlashSystem storage.

A key feature is that up to 40 terabytes (TB) of CAPI-attached flash storage may be employed as an alternative to RAM. According to the developers, a single POWER8-based system may handle workloads that would otherwise require twenty-four x86 servers. BigRedis exploits POWER8 SMT at up to eight threads per core.



Appendix: OpenPOWER Foundation

Overview The OpenPOWER Foundation was formed in August 2013 by Google, IBM, Mellanox, NVIDIA and Tyan, and has since expanded to include 100+ members. Many of these have cooperated with IBM, and with each other on supercomputing as well as commercial projects.

The OpenPOWER Foundation currently offers three levels of paid membership – Platinum, Gold and Silver – with varying fees and levels of representation on its Board of Directors. There is also a no-charge category of Associate (individual) and Academic (institutional) membership.

Members may participate in Working Groups that define OpenPOWER standards in specific areas. Current groups address hardware architecture; accelerator, memory and physical interfaces; open source systems and applications software; development platforms; and compliance. The chair of each Working Group and representatives of all Platinum Members form the OpenPOWER Foundation’s Technical Steering Committee.

In March 2015, the OpenPOWER Foundation announced the formation of an Advisory Group whose members include the Linux Foundation; Facebook’s Open Compute Project; and the China POWER Technology Alliance, a government-sponsored umbrella group for Chinese participants in the OpenPOWER Foundation.

Members and Activities Vendors and academic institutions participating in the OpenPOWER Foundation, and their POWER8- and CAPI-related activities are summarized in the table below. Data is current as of March 2015.

As membership continues to expand, and new projects are launched, readers should check http://openpowerfoundation.org/membership/current-members/ for the latest information.

PLATINUM Members Altera (US) Company information: Leading supplier of FPGA-, CPLD- (complex programmable logic device) & SOC- (system on a chip) based accelerators, design & development tools for high-performance computing (HPC) & data center applications.

POWER8 plans: FPGA-based acceleration platform supports Intel OpenCL. Company claims FPGA- accelerated systems can achieve 5-20x higher performance than standard CPU-based servers. Altera FPGAs are used in IBM Data Engine for NoSQL & IBM Data Engine for Analytics. Co-developed OpenPOWER CAPI Development Kit with IBM & Nallatech.

Canonical (US) Developer of Ubuntu Linux distribution & other open source software.

POWER8 plans: Ubuntu Server for POWER8, TurboLAMP stack developed with IBM, MariaDB, Mellanox Technologies & Zend.

Google (US) Leading supplier of search engine, e-commerce, cloud, application, e-mail & other Internet services. Company is believed to operate between one & two million servers at 30+ locations worldwide.

POWER8 plans: Google has developed a two-socket Power8-based motherboard for internal testing & use. Company is reported to be interested in the larger number of virtual machines (VMs) compared to x86 architecture.

Mellanox Technologies (Israel/US) Company information: Leading supplier of InfiniBand & Ethernet interconnects for servers & storage. Key IBM partner for Big Data analytics, cloud & software defined Elastic Storage solutions.

POWER8 plans: Power8 platform incorporates Mellanox 40 Gb Ethernet networking gear & TurboLAMP stack co-developed with Canonical, IBM, MariaDB & Zend. In March 2015, the company announced CAPI integration of new ConnectX-4 EDR programmable adapter supporting 100 Gb/s InfiniBand & Ethernet. Company’s EDR 100 Gb solutions will also be employed in new supercomputer project awarded to IBM & NVIDIA for DOE Oak Ridge Lawrence Livermore National Laboratories.



PLATINUM Members (cont)

Micron Technology (US) Company information: Leading supplier of RAM & flash memory devices.

POWER8 plans: Established IBM development partner for flash memory used in IBM FlashSystem storage array. Most recent cooperation has been in integration of Micron FortisFlash NAND & IBM PowerCore optimization technology for use in IBM FlashSystem 900 & 9000, introduced February 2015.

NVIDIA (US) Company information: Leading supplier of GPUs, other semiconductor products & software for HPC- & other graphics- & image-intensive applications. Claims that more than 85% of accelerated HPC systems (supercomputers) employ its technology.

POWER8 plans: Offers Tesla Accelerated Computing Platform – GPUs & enabling software – for POWER8, along NVLink interconnect & CUDA parallel processing development environment. NVLink – a higher-speed (NVIDIA claims more than 5x) alternative to PCIe – will enable NVIDIA GPUs to access the full bandwidth of POWER8 CPU memory. NVLink will also be licensed to other OpenPOWER Foundation members, & will be incorporated in a future IBM POWER processor design. Also offers PGI optimizing Fortran, C & C++ compilers for POWER8-based systems. According to company, these provide a user interface, language parallel programming & optimization features identical to those on PGI x86 Linux compilers. Company expects most GPU-accelerated x86 applications currently built with PGI compilers will port to GPU-accelerated POWER8 systems with simple recompile. IBM, NVIDIA & others are developing next-generation supercomputers under a $425 million contract from the US DOE.

Samsung (ROK) Company information: Diversified global manufacturer of semiconductor, consumer electronics & appliances, & industrial electronics products. Leading manufacturer of ARM processors.

POWER8 plans: None announced.

SK Hynix (ROK) Company information: Global semiconductor manufacturer. POWER8 plans: None announced.

Suzhou PowerCore Technology Company (China) Company information: Start-up affiliated with Research Institute of Jiangsu Industrial Technology, Jiangsu province.

POWER8 plans: Provides customized POWER8-based processor designs for Chinese server manufacturers. In March 2015, the company announced CP1 processor for use by these.

TYAN (subsidiary of Mitac Computing Technology Corporation) (Taiwan) Company information: Designs, manufactures & markets high-volume x86-based servers & workstations.

POWER8 plans: Began delivery of low-cost, single-socket POWER8-based Customer Reference System server in November 2014. Supplies this product to Google. In March 2015, the company announced TN71-BP012, the first commercially available POWER8-based server. GA is 2Q15. Will be used by IBM for new SoftLayer bare metal service.

GOLD Members Avnet (US) Company information: Distributor of components, computers, storage & software from numerous vendors. Also offers value-added services. Believed to be world’s largest IT distributor.

POWER8 plans: Markets, supports & offers integration services for IBM POWER8-based hardware & software to vendors as well as end users. Cooperates with IBM as well as other OpenPOWER Foundation members. For POWER8-based systems, company focuses on analytics & other high-performance applications.

Beijing Teamsun Technology Company (China) Company information: IT services company operating in Greater China & Southeast Asia.


Chuanghe Century Telecom (China) Company information: Diversified supplier to the Chinese telecommunications industry.

POWER8 plans: Announced China-branded POWER8-based servers will be available in 2015.

Hitachi (Japan) Company information: Diversified global manufacturer of electronics, consumer & industrial projects. Leading semiconductor manufacturer.




GOLD Members (cont) Inspur (China) Company information: Diversified supplier of servers, mass storage, software & services for government & commercial customers in China.

POWER8 plans: Will use OpenPOWER Foundation reference design to build new servers.

Research Institute of Jiangsu Industrial Technology (UNIS) (China) Company information: Affiliate of Jiangsu University of Science & Technology charged with developing technology in Jiangsu province. Cooperates with Suzhou PowerCore.

POWER8 plans: Supports development of POWER8-based solutions by members of the China POWER Technology Alliance led by the China Ministry of Industry & Information Technology (MIIT).

Sichuan Huaxun Zhongxing Technologies (China) Company information: Joint venture with IBM for development & commercialization of Smart Cities & Internet of Things (IOT) solutions in Sichuan province. Sponsored by provincial government.

POWER8 plans: Companies will extend cooperation to develop mainframe-class servers based on IBM POWER8 & OpenPOWER Foundation technologies.

VeriSilicon (China) Company information: Supplier of custom hardware & software designs to semiconductor companies, OEMs, & large Internet platform companies. Claims 200+ customers worldwide.

POWER8 plans: Will offer design services to suppliers of POWER8-based systems.

Wistron (Taiwan) Company information: Major contract manufacturer of servers, storage, PCs, LCD TVs, mobile devices & other products. Provides custom design, manufacturing & after-sales services.

POWER8 plans: Dual-socket POWER8-based design contains IBM & Wistron hardware technologies. Servers based on this design will be manufactured by Wistron & marketed by IBM & Wistron. In March 2015, the company demonstrated prototype of POWER8-based HPC system including NVIDIA & Mellanox components.

Zoom Networks (China) Company information: Provides system integration, network design & construction, software development, & other services to telecom operators in China.

POWER8 plans: In March 2015 announced RedPower servers using PowerCore CP1 processor. Servers will come to market in 2015.

ZTE (China) Company information: Supplier of systems, equipment & services for the Chinese telecommunications industry.


SILVER Members Algo-Logic Systems (US) Company information: Specializes in FGPA-based acceleration solutions for data center search, ultra low latency finance (real-time trading), & IOT real-time processing.

POWER8 plans: Will support POWER8- & Altera- & Xilinx FGPA-based solutions for accelerated finance & key value search applications.

Bluebee (Netherlands) Company information: Company’s Genome Analytics Platform combines advanced cancer detection & treatment algorithms with hybrid-core architecture. Company has cross-marketing agreement with Convey Computer Corporation, also an Open Software Foundation (OSF) member.

POWER8 plans: Supports POWER8 & CAPI.

Bull (subsidiary of Atos SE) (France) Company information: Systems vendor offering Escala servers (IBM Power OEM systems), mainframes, x86 rack & blade servers, & HPC systems. Recently acquired by IT services company Atos SE (Societas Europaea), with 80,000+ staff worldwide. Both are longstanding IBM partners.

POWER8 plans: Markets OEM Power8-based Escala servers with Power Linux & AIX. Supports CAPI & OpenPOWER Foundation technologies.

Celestica (Canada) Company information: Contract manufacturer of computer, electronics, mobile devices & other computer & communications products. Provides custom design, prototyping, assembly, testing, product assurance & logistics services worldwide

POWER8 plans: None announced

Chelsio Communications (US) Company information: Leading supplier of ASIC-based PCIe Gen 3 adapters enabling users to offload processing of TCP, UDP, iWARP, iSCSI, FCoE & other common protocol stacks from servers & storage systems. Reports 100+ OEM customers.

POWER8 plans: Will offer Unified Wire Adapters based on company’s Terminator (T5) ASIC technology for POWER8- & CAPI-based systems.



SILVER Members (cont) Cirrascale (US) Company information: Designs & manufactures energy-efficient blade-based compute & storage servers & infrastructure for cloud, conventional & modular data centers.

POWER8 plans: In March 2015, the company announced RM4950 Development Server supporting POWER8 & up to four NVIDIA Tesla GPUs. Co-developed with NVIDIA & Tyan. System is intended for “early experimentation & product or code development” rather than production use. GA is 2Q15.

Convey Computer Corporation (US) Company information: Supplier of FPGA-based accelerators used in life sciences, engineering/scientific, big data, government/military, seismic processing & other applications.

POWER8 plans: In March 2015, the company CAPI Developer Kit using POWER8 CPU, & Xilinix & Altera FPGA-based co-processors.

DataDirect Networks (US) Company information: Supplier of massively scalable storage systems that are widely employed among “Top 100” supercomputer users worldwide.

POWER8 plans: Supports POWER8 & OpenPOWER Foundation offerings for supercomputing, life sciences & genomics, seismic processing, financial services trading & risk analysis, multimedia, video surveillance & other applications.

Emulex (US) Company information: Leading supplier of I/O devices, including OEM high-performance Ethernet & Fibre Channel-based products designed into numerous server & storage platforms.

POWER8 plans: Expects to provide Gen 5 Fibre Channel & 40 Gb Ethernet solutions for OpenPOWER Foundation designs.

Exablaze (Australia) Company information: Supplies FPGA-based accelerators, including network interface cards, switches & software primarily for real-time trading operations in large banks. Company is seeking to expand into HPC, data center, telecommunications & other low-latency markets.

POWER8 plans: Will integrate ExaNIC cards into POWER8 & CAPI ecosystem & pursue new opportunities with OpenPOWER Foundation members.

Fusion-io (subsidiary of SANdisk Company) (US) Company information: Leading supplier of flash memory-based PCIe accelerators for servers & storage. Markets directly & through OEM customers including Cisco Systems, Dell, IBM, HP, Lenovo & Supermicro. Claims that its products are used by over 75% of Fortune 100 companies.


Global Intelligence Solutions (GIS) Federal (US) Company information: Specialized supplier of geospatial engineering, big data analytics & visualization systems & software. Main business is with military & intelligence agencies, but also serves other government & commercial clients.


HGST (subsidiary of Western Digital) (US) Company information: Leading manufacturer of hard disk drives & SSDs. Formerly Hitachi Global Storage Technologies.


Inphi (US) Company information: Leading supplier of ASIC- & FPGA-based devices for high-speed analog & mixed signal processing in network & data center architectures. Company designs & develops products for servers, switches, routers, storage & other computer & communications equipment.


Interface Masters Technologies (US) Company information: Leading supplier of server adapters, NICs, embedded switches & other products for monitoring & management of Gigabit, 10 Gigabit & 40 Gigabit Ethernet networks.


Inventec (Taiwan) Company information: OEM manufacturer of notebook computers, servers & mobile devices. Has also entered cloud computing, mobile computing, network applications, digital home appliances & sustainable energy markets.


KNS Group (India) Company information: Privately held Indian conglomerate. POWER8 plans: None announced.



SILVER Members (cont) Lawrence Livermore National Laboratory (US) Company information: National laboratory operated by US DOE. Primary responsibility is nuclear weapons research.

POWER8 plans: 1114 DOE awarded IBM, NVIDIA, Mellanox & others contracts valued at $425 million to deliver two Open Power-based supercomputers to Lawrence Livermore & Oak Ridge National Laboratories, Systems will include “thousands” of next-generation POWER & NVIDIA Volta processors, & are expected to deliver c. 150 petaflops performance when completed in 2017-18.

Maxeler Technologies (US) Company information: Supplier of FPGA-based accelerators, system appliances & software for HPC applications in oil & gas, financial trading, scientific computing, & other industries. Company employs distinctive dataflow architecture. Hardware includes Intel Xeon processors & Xilinx FPGAs.

POWER8 plans: Has conducted extensive internal testing of POWER8, CAPI & other OpenPOWER Foundation technologies, but no plans announced to date.

Beijing Memblaze Technology Company (China) Company information: Supplier of high-performance PCIe flash memory cards.


Nallatech (subsidiary of ISI) (US) Company information: Leading supplier of FPGA-based accelerators. Subsidiary of Interconnect Systems Inc. (ISI), a contract designer & manufacturer of miniaturized electronics systems.

POWER8 plans: Offers OpenPOWER CAPI Developer Kit System enables Nallatech 385 to attach to IBM POWER8 processors via CAPI. Said to be first CAPI-supported FPGA accelerator card. Co-developed with Altera & IBM. Nallatech 385 architecture said to be “ideally suited for algorithm-acceleration applications” including Monte Carlo simulations, key-value stores, & financial & medical algorithms.

Nanjing Byosoft Company (China) Company information: Develops & integrates BIOS technology for computers, embedded systems, communications & other products. Supports x86 & ARM standards. Currently targeting IOT applications. Claims to be only BIOS company in China.

POWER8 plans: Will support POWER8, CAPI & other OpenPOWER Foundation standards.

Numascale (Norway) Company information: Supplier of clustered HPC systems & technology used in engineering & scientific computation. Customers include Norwegian state oil company Statoil as well as supercomputing centers worldwide.


OCF (UK) Company information: Systems integrator specializing in custom HPC server & storage clusters. Offers hardware & software, professional services & cloud delivery. Customers are primarily U.K academic institutions, public sector agencies & private companies. Previous IBM channels partner.


One Stop Systems (US) Company information: Supplier of high-density GPU appliances, & flash storage & disk arrays.


OVH (France) Company information: ISP operating primarily in Europe, but expanding into North America. Claims 700,000+ customers with 180,000+ hosted environments in 17 data centers worldwide.

POWER8 plans: New RunAbove service offers access to IBM POWER8 server capacity with KVM, OpenStack, & Fedora or Ubuntu for development. Company reports that internal testing shows 3x higher performance than equivalent x86-based architectures.

PMC (US) Company information: Fabless semiconductor company providing communications & storage devices for OEMs including Alcatel-Lucent, Cisco, EMC, Fujitsu, Hitachi, Huawei, HP, Juniper, Mitsubishi & ZTE. Outsources wafer fabrication & assembly. Offers 250+ different products.

POWER8 plans: Company will work with IBM & other OpenPOWER Foundation members to develop server & storage solutions for next-generation data centers that integrate IBM POWER CPUs, PMC SAS storage & SSDs.

QLogic (US) Company information: Leading supplier of Fibre Channel, Ethernet & iSCSI adapters for servers & storage systems. Competes directly with Emulex, also an OpenPOWER Foundation member.

POWER8 plans: Will support POWER8, CAPI & other OpenPOWER Foundation standards for current & next-generation adapters.



SILVER Members (cont) Rackspace (US) Company information: Online hosting company reports more than 200,000 customers & 100,000 servers worldwide – some customers employ multiple servers. Has cooperated with key OpenPOWER Foundation members since mid-2012 with the stated objective of reducing company’s dependence on Intel technology. Co-founder of OpenStack consortium.

POWER8 plans: Developing an open server design leveraging technologies developed by the Open Compute developers' community & OpenPOWER Foundation. Company’s design will employ IBM Power8 processors, along with components sourced from other parties. In March 2015, the company demonstrated prototype motherboard for this system.

Redis Labs (US) Company information: Company is the leading commercial supplier of software & services for open source NoSQL database Redis. Claims more than 4,000 paying customers, primarily accessing Redis cloud services for development purposes. Supports Redis community. Available on IBM SoftLayer.

POWER8 plans: Cooperates with IBM & other OpenPOWER Foundation members for POWER8 & CAPI Redis enablement. Developed IBM Data Engine for NoSQL, along with IBM, Altera & Canonical. This appliance solution includes Redis Labs software, Ubuntu, Altera FPGAs, & IBM Power8-based systems, CAPI & flash storage.

RIKOR Information Systems (Russia) Company information: Designs & manufactures energy-efficient high-density servers. Also plans to build storage & network devices.


RTDS Technologies (Canada) Company information: Supplier of digital power system simulators used by utilities worldwide.

POWER8 plans: Will support POWER8, CAPI & other OpenPOWER Foundation technologies in future product offerings.

Sanmarco Informatica (Italy) Company information: Subsidiary of ERP software vendor Jgalileo POWER8 plans Will offer open source ERP solutions for POWER8-

based systems.

Servergy (US) Company information: Specialist supplier of low energy consumption servers.

POWER8 plans: Will incorporate POWER8, CAPI & other OpenPOWER Foundation technologies in future product offerings.

StackVelocity (business unit within Jabil Circuit) (US) Company information: Specializes in OEM design & production of HPC systems for hyperscale datacenter operators.

POWER8 plans: Will provide OEM POWER8- & CAPI-based HPC systems to OpenPOWER Foundation members.

Synapse Design (US) Company information: Provides contract ASIC & System-on-a-Chip (SoC) design services.

POWER8 plans: Will provide design services for POWER8, CAPI & other OpenPOWER Foundation technologies.

Xilinx (US) Company information: Leading supplier of FGPAs, CPLDs & other programmable devices for wide range of clients worldwide. Fabless; i.e., outsources wafer fabrication & assembly.

POWER8 plans: Wiil support POWER8, CAPI & other OpenPOWER Foundation technologies for current as well as future products. Claims internal tests show significantly higher key value search performance – up to 36x – with CAPI-attached FGPAs & flash memory, compared to equivalent x86 configurations.

ACADEMIC Members Cineca (Italy) Consortium of 70+ Italian universities & research institutions under the aegis of the country’s Ministry of Education. Operates the largest computer center in Italy, providing HPC, network & Web-based services for members. Longstanding IBM Blue Gene user & development partner. Also NVIDIA partner.

Coimbatore Institute of Technology (India) Research Institute affiliated with the Anna University of Technology. Has operated IBM Center of Excellence for HPC since 2009. IBM & NVIDIA development & education partner in supercomputing.

Hartree Center – Science & Technology Facilities Council (UK) UK government agency supporting development in key scientific fields, including particle physics, astronomy/space science & nuclear physics Funds programs at 50+ universities & research institutes. Operates supercomputing infrastructure supporting these. IBM & NVIDIA development partner.

Indian Institute of Technology – Bombay (India) Government-supported research & education institution active in medical, aerospace & IT, including HPC systems. Longstanding development partner of IBM India Software Lab.

Indian Institute of Technology – Roorkee (India) Research & education institution funded by state government of Karnataka & local IT industry partners. IBM supports development in supercomputing, information management & data mining.



ACADEMIC Members (cont) Institute of Communication & Computer Systems (Greece) Academic group managed by country’s Ministry of Advanced Technology areas. Active in supercomputing research using IBM & other vendors’ HPC systems. Interaction Australasia (Australia) Industry group for IBM systems users in Australia, New Zealand other Asia/Pacific countries. Focuses mainly on IBM Power Systems & i. Affiliated with COMMON USA & COMMON Europe. International Institute of Information Technology Bangalore (India) Indian Deemed University funded by state government of Karnataka & IT industry partners. IBM & NVIDIA support supercomputing development & education. Jülich Supercomputing Centre (Germany) Power Acceleration & Design Center in Munich, Germany conducts supercomputer system & software R&D development around IBM POWER8 & NVIDIA Tesla GPUs. Center draws on existing partnerships with NVIDIA & IBM – has used & cooperated with IBM through multiple generations of BlueGene systems. Center will formalize & enhance collaboration that the parties have already been engaged in.

Oak Ridge National Laboratory Lab (US) IBM & NVIDIA are collaborating under a $425 million contract from the US DOE to develop supercomputer systems for the department’s Oak Ridge & Lawrence Livermore National Laboratories. Systems will include “thousands” of next-generation POWER & NVIDIA Volta processors, & are expected to deliver c. 150 petaflops performance when completed in 2017-18.

Oregon State University (US) Open Source Lab provides hosting for 160+ open source development projects involving Linux, Apache, KVM, OpenStack & other components. Services now include Power Linux & OpenPOWER development & testing. Lab offers POWER7+ LPAR-based big endian instances & POWER8-based big or little endian instances running on KVM & providing access via OpenStack API & GUI interface.

Pohang University of Science & Technology (POSTECH) (ROK) Private university’s Information Research Laboratory conducts R&D in HPC systems & provides supercomputing support for biotechnology, nanomaterials, engineering & compute-intensive disciplines. Operates two centers in cooperation with the German Max Planck Society.

Rechenzentrum Garching – Max Planck (RZG) (Germany) Provides computing & communications services for 20+ German research institutes affiliated with the Max Planck Society. Major supercomputing & mass storage center supports research in astrophysics, fusion research, materials sciences, biochemistry, polymers & other areas. Longstanding IBM & NVIDIA development partner.

Rice University (US) Research Computing Support Group provides supercomputing systems supporting research by university staff & affiliates in genomics & other areas. Longstanding user of IBM Blue Gene & customized HPC systems. Recent initiatives, in collaboration with IBM, NVIDIA & Mellanox, focus on the application big data & analytics technologies in medical research & clinical practice.

Sandia National Laboratories (US) US DOE R&D center dealing with nuclear weapons. Also conducts research in computational biology, mathematics, materials science, alternative energy, hazardous waste disposal & other areas. Longstanding user of IBM supercomputers, including systems such as ASCII Red & more recently ASCII Red Storm. Close development partner with IBM & NVIDIA.

Seoul National University (ROK) Public university conducts research on, & provides services to faculty, students & affiliates employing HPC systems. Supercomputers include IBM Linux clusters, internally developed & other systems. Major research focus on GPU-based clusters & software.

Shanghai Jiao Tong University (China) Key university administered by country’s Ministry of Education. Operates High Performance Computing Center (HPCC) supporting wide range of research projects. Supercomputing development partner of IBM & other vendors. NVIDIA CUDA Center of Excellence (CCOE) since 2011.

Symbiosis Institute of Computer Studies & Research (India) Leading IT Business School at Symbiosis International University. IBM supports development & education in business analytics, cognitive computing & other areas.

Tsinghua University (China) Leading university orates supercomputers supporting wide range of scientific & engineering applications. Main platform is currently Intel-based system supplied by Inspur, also an OpenPOWER Foundation member. NVIDIA CUDA Center of Excellence since 2009. Will focus on POWER8-based earth sciences applications.

University of Arkansas High Performance Computing Center (AHPCC) (US) Supports supercomputing research in computer science, nanoscience, computational chemistry & biomagnetics, materials science & spatial science. Center is funded by US National Science Foundation & Arkansas Science &Technology Authority. Longstanding IBM HPC user & development partner launched IBM iDataplex-based supercomputer in 2011. Developing POWER8-based hybrid cluster with NVIDIA GPUs.



ACADEMIC Members (cont) University of Campinas (UNICAMP) (Brazil) Institute of Computing provides education in computer science, including HPC systems. Delivers IT services to users in all major university disciplines. Also manages National Center for High Performance Computing (CENAPAD-SP) in São Paulo, funded by country’s Ministry of Science & Technology.

University of Central Florida (US) Advanced Research Computing Center (ARCC) provides high performance computing (HPC) resources for research by faculty & students. Disciplines include astronomy, biology, chemistry, engineering, modeling & simulation, nanoscience & optics. Longstanding IBM & NVIDIA partner for development & education.

University of Hyderabad (India) Department of Computer Science of leading Indian research university provides supercomputing services to internal users as well as affiliates in medicine, economics, engineering, energy, neural & cognitive sciences, & other disciplines. Longstanding IBM development partner for POWER-based systems, high-end storage & software.

University of Illinois (US) National Center for Supercomputing Applications at Urbana-Champaign conducts wide range of HPC projects. Center is supported by National Science Foundation, other US federal agencies, State of Illinois & industry partners.

University of Oregon (US) NSF-funded High Performance Computing Cluster, ACISS, provides scientific supercomputing support for internal as well as affiliated users. Center is a longstanding IBM development partner, also works with NVIDIA.

University of Patras (Greece) University’s HPC Lab provides supercomputing services for numerous Greek public sector clients. Has cooperated with IBM, NVIDIA & other vendors on development & education programs.

Waseda University (Japan) Private university conducts research in system, software & network design; artificial intelligence & robotics; cognitive computing & other advanced technology areas. Department of Computer Science & Engineering researches HPC systems, & operates Fujitsu supercomputers supporting university users & affiliates.



International Technology Group

ITG sharpens your awareness of what’s happening and your competitive edge . . . this could affect your future growth and profit prospects

International Technology Group (ITG), established in 1983, is an independent research and management consulting firm specializing in information technology (IT) investment strategy, cost/benefit metrics, infrastructure studies, deployment tactics, business alignment and financial analysis.

ITG was an early innovator and pioneer in developing total cost of ownership (TCO) and return on investment (ROI) processes and methodologies. In 2004, the firm received a Decade of Education Award from the Information Technology Financial Management Association (ITFMA), the leading professional association dedicated to education and advancement of financial management practices in end-user IT organizations.

Client services are designed to provide factual data and reliable documentation to assist in the decision-making process. Information provided establishes the basis for developing tactical and strategic plans. Important developments are analyzed and practical guidance is offered on the most effective ways to respond to changes that may impact complex IT deployment agendas. A broad range of services is offered, furnishing clients with the information necessary to complement their internal capabilities and resources.

Clients include a cross section of IT end users in the private and public sectors representing multinational corporations, industrial companies, financial institutions, service organizations, educational institutions, federal and state government agencies as well as IT system suppliers, software vendors and service firms. Federal government clients have included agencies within the Department of Defense (e.g., DISA), Department of Transportation (e.g., FAA) and Department of Treasury (e.g., US Mint).

Copyright © 2015 International Technology Group. All rights reserved. Material, in whole or part, contained in this document may not be reproduced or distributed by any means or in any form, including original, without the prior written permission of the International Technology Group (ITG). Information has been obtained from sources assumed to be reliable and reflects conclusions at the time. This document was developed with International Business Machines Corporation (IBM) funding. Although the document may utilize publicly available material from various sources, including IBM, it does not necessarily reflect the positions of such sources on the issues addressed in this document. Material contained and conclusions presented in this document are subject to change without notice. All warranties as to the accuracy, completeness or adequacy of such material are disclaimed. There shall be no liability for errors, omissions or inadequacies in the material contained in this document or for interpretations thereof. Trademarks included in this document are the property of their respective owners.

itg1505 power8 linux migration - digitaltrend technology group may 2015 strategic considerations for...

Documents