IBM Linux ONE Emperor II Technical Overview Your

Topics for Today q. IBM Linux. ONE Emperor™ II Key Messages & Value Propositions q. IBM Linux. ONE Emperor II System Overview v. Processor improvements v. Pause-less Garbage Collection v. Pervasive Encryption v. IBM Secure Service Container q. Hypervisor & Operating System Support q. Associated Availability Dates 3 © 2017 IBM Corporation

IBM Linux. ONE Emperor II Key Messages and Value Propositions q. Key messages: v Industry leading security without re-engineering, encrypting all in-flight and at-rest data v Accelerating the data that drives your business, unprecedented data serving performance v Scale that maps to your growth vision, flexible, granular, powerful scalability options q. Value propositions v x 86 data centers have a new, more secure alternative with the pervasive encryption of IBM Linux. ONE Emperor II v Data serving at a new scale, with faster cores than x 86, more cache-per-core, large memory (up to 32 TB) to support in-memory databases and analytics, and up to 832 GB/S total I/O bandwidth to let you run billions of transactions per day on a single Emperor II v Open & flexible – allowing you to mix commercial data serving & analytics with open source data serving & analytics on the same system without compromising SLAs or security v Enhanced Java affinity – Emperor II runs Java™ faster than x 86 does, and has exclusive hardwarebased Pause-Less Garbage Collection to provide more predictable and consistent response times for Java apps 4 © 2017 IBM Corporation

Linux. ONE Emperor II System Overview IBM and BP Confidential – This material is

Quick View of IBM Linux. ONE Emperor II q Physical organization is cores on chips, chips & memory in up to 4 CPC (Central Processor Complex) drawers, PCIe in I/O drawers, drawers in a cabinet, dualcabinet system q Five models: LM 1 (1 CPC 5 -socket drawer), LM 2 (2 CPC 5 -socket drawers), LM 3 (3 CPC 5 -socket drawers), LM 4 (4 CPC 5 -socket drawers), LM 5 (4 CPC 6 -socket drawers) q Cores (Processor Units or PUs) v Can be either Integrated Facility for Linux cores (IFLs), System Assist Processors (SAPs), an Integrated Firmware Processor (IFP), or spares v 5. 2 GHz, in-core crypto & compression, SMT-2 v 41 PUs per processor drawer on models LM 1 -LM 4; LM 5 has 49 PUs per drawer § 5 cores per drawer are reserved as SAPs (+3 more on LM 5), which assist with disk I/O operations § Remaining 36 cores can be configured as any mix of IFLs, SAPs, and spares § 1 core/system is reserved as the IFP, which assists with PCIe operations for Ro. CE and z. EDC § 2 core/system are reserved as spares § The mandatory 2 spares and the IFP are allocated in the 1 st drawer, leaving up to 33 IFLs in LM 1 v Max Linux cores per model: 33 (36 - IFP & 2 spares), 69 (33+36), 105 (69+36), 141 (105+36), 170 v 1 optional CP available solely to support the GDPS Virtual Appliance v Up to 85 LPARs (Logical Partitions or “Hard Partitions” as viewed by licensed software) q Memory v High availability Redundant Array of Independent Memory (RAIM) v Minimum of 64 GB user memory + 192 GB Hardware System Area (HSA) = 256 GB system minimum v Maximum of up to 8 TB/drawer, 32 TB total for System; up to 16 TB per LPAR (OS/hypervisor dependent) q I/O v A single System Control chip serves each drawer by interconnecting the processor, memory and IO, providing 672 MB of L 4 cache accessible to all cores in the entire system v New FICON® (Fibre Connection), OSA (Open Systems Adapter), and Ro. CE (RDMA over Converged Ethernet) cards v 16 GBps PCIe Gen 3 I/O interconnects 6 © 2017 IBM Corporation IBM Linux. ONE Emperor II Machine Type: 3906 Models: LM 1, LM 2, LM 3, LM 4, Max # of Max Model LM 5 Linux Cores Memory LM 5 170 32 TB LM 4 141 32 TB LM 3 105 24 TB LM 2 69 16 TB LM 1 33 8 TB*

System Design Changes q 14 nm Processor with improved SIMD, SMT q 10 Cores per CP SCM design q 5 (on LM 1 -LM 4) or 6 (on LM 5) CP SCMs per Drawer q Integrated I/O with PCIe Direct Attach q Now only one System Controller (SC) chip per drawer q Crypto Express 6 S q OSA-Express 6 S IBM Linux. ONE Emperor II q Simplified CPC Drawer SMP Fabric 7 © 2017 IBM Corporation q FICON Express 16 S+ q Ro. CE Express 2 q Radiator Design improvements q Expanded operating environment (ASHRAE Class A 3)

Linux. ONE Emperor II – Highly Engineered System for Data Serving Up to 170 Linux cores running at 5. 2 GHz in an Emperor II (3906) Plus up to 24 cores for offload system processing Plus up to 4, 480 dedicated I/O processors Up to 320 pairs of Power. PC® cores and… Up to 23 SAPs 1 IFP Up to 320 dozens of programmable embedded engines performing I/O Share up to 170 processors across up to 85 LPARS; plus minimum 2 spares LM 1: 33 cores, 8 TB LM 2: 69 cores, 16 TB LM 3: 105 cores, 24 TB LM 4: 141 cores, 32 TB LM 5: 170 cores, 32 TB 8 System Assist Processors (SAPs) are multithreaded cores that orchestrate disk I/O on behalf of the Linux cores, and do not factor into software license core counts 1 Integrated Firmware Processor (IFP), which is used for Ro. CE Express and z EDC Express I/O cards. These cores do not factor into software license core counts Up to 320 pairs of Power. PC® cores across 160 I/O adapters, custom data-moving hardware & firmware, and a “swarm” (up to 3, 840) of small lightweight processors that make the I/O channels fast and effective at moving I/O data to and from memory © 2017 IBM Corporation

Processor Improvements IBM and BP Confidential – This material is for educational purpose only

Linux. ONE Processor Design Summary q q q 10 14 nm SOI technology 17 layers of metal 6. 1 billion transistors 10 cores per chip 5. 2 GHz q Cache Improvements – New power efficient logical directory design – 33% Larger L 1 I$ cache (128 K vs. 96 K on Emperor I) – 2 x Larger L 2 D$ cache (4 MB vs. 2 MB on Emperor I) – 2 x Larger L 3 cache (128 MB vs. 64 MB on Emperor I) with symbol ECC q New Translation/TLB 2 design – 4 concurrent translations – Reduced latency – Lookup integrated into L 2 access pipe – 2 x CRSTE growth – 1. 5 X PTE growth – New 64 entry 2 gig TLB 2 q Pipeline Optimizations – Improved instruction delivery – Faster branch wakeup – Reduced execution latency – Improved OSC* avoidance – Optimized 2 nd generation SMT 2 q Better Branch Prediction – 33% Larger Branch Target Buffer 1 & BTB 2 – New Perceptron Predictor – New Simple Call Return Stack © 2017 IBM Corporation

Linux. ONE Processor Design Improvements q Performance Enhancements – Larger L 1, L 2, and L 3 caches – New address translation design – Branch prediction innovations (Linux® aware) – Instruction pipeline optimizations – Increased floating-point compute bandwidth – 2 nd generation SMT (the SAPs can now use SMT) q Architecture features – SIMD extensions for decimal operations and analytics 14 nm SOI Technology Up to 10 active cores (PUs) per chip • Register-based BCD operations – Guarded Storage Facility • Enables pause-less garbage collection for Java – Instruction Execution Protection Facility • “No-execute” zones – help reduce impact of garbage collection q Crypto co-processor System Control (SC) chip 11 © 2017 IBM Corporation

z 14 Processor Enhancements: Guarded Storage Facility (GSF) for Pause-less Garbage Collection q. Problem: v. When garbage collection occurs today, all threads running under a JVM must stop v. Customers are consolidating from multiple to single JVM environments to increase productivity and save money. v. The consolidation effort generates heap sizes >100 GB where garbage collection pauses can take minutes! v. Long pause times cause transactional application failures and SLA violations. q. Solution: v. Define flexible new architecture that provides hardware assisted read barriers for guarded storage involved in a garbage collection/compaction event. v. Whenever a Pointer is loaded from memory, the pointer is checked against a pending GC, and in case of a “hit”, the control flow is redirected v. The Dynamic Runtime can then assist in GC-ing the pointed-to object, before resuming the SW thread. v. Software exploitation of fast hardware barrier detection and acceleration will allow application threads to run concurrently during the majority of garbage collection events q. Impact: v. Reduces worst case latency impacts for critical applications like financial trading platforms v. Maintains SLAs, and keeps IBM Z® servers in our customers modernization roadmaps. 12 © 2017 IBM Corporation

Pause-less Garbage Collection Normal GC pause q. GC is the reclaiming of memory that is no longer used programs v Normal GC pauses all threads while it does its work v Pause-less GC allows Java GC to run alongside with applications, which means response times are more predictable Pause-less GC v Pause-less GC also helps ensure high-performance when transaction processing reaches high volumes q. Java Garbage Collection - Tuning made easy q. Up to 3 x better throughput for response time-constrained Service Level Agreements (SLAs) q. Up to 10 x better average GC pause times IBM Java 8 SR 5 + IBM Linux. ONE™ Guarded Storage Facility allows GC to run in parallel with application. Enable with: -Xgc: concurrent. Scavenge 13 © 2017 IBM Corporation

Co-processor Compression (CMPSC) Improvements Dedicated Co-Processor per Core § With small, dedicated dictionary cache § z 196 z. EC 12 Co-Processor § Compression § Expansion § Translation 14 Connects to main cache-structure (better throughput for many dictionaries) § z 13 § New compression algorithm (fully pipelined, ~2 x faster) Huffman compression on top of CMPSC compression (Embedded in existing dictionary, reuse of generators) § z 14 Order Preserving compression in B-Trees and other index structures § Faster expansion algorithms § Reduced overhead on short data § Compression/Expansion Strategy (IBM z 14™ and beyond) • Improved compression/expansion performance, in particular reduced overhead for short compression disk/memory savings by exploiting more compression with low CPU consumption • Improved compression ratio (e. g. Huffman coding in z 14) additional disk/memory savings even where compression is already in use today • Order-preserving compression for search trees, sort files, etc. disk/memory savings for large parts of data that have not been practical to compress previously © 2017 IBM Corporation

Compression Coprocessor (CMPSC) vs. z. EDC Using the right hardware compression acceleration for each of your workloads Compression Coprocessor z Enterprise Data Compression On Chip PCIe Adapter In every IBM Z server Introduced with IBM z. Enterprise® EC 12 GA 2 and IBM z. Enterprise BC 12 Mature: Decades of use by Access Methods and Db 2® Work is performed jointly by CPU and Coprocessor Proprietary Compression Format Mature: Industry Standard with decades of software support Work is performed by the PCIe Adapter Standards Compliant (RFC 1951) Use Cases Small object compression § Rows in a database Large Sequential Data § QSAM/BSAM Online Sequential Data Industry Standard Data § Cross Platform Data Exchange § Objects stored in a data base Users § VSAM for better disk utilization § Db 2 for lower memory usage § The majority of customers are currently compressing their Db 2 rows 15 Users § QSAM/BSAM for better disk utilization and batch elapsed time improvements § Java for high throughput standard compression via java. util. zip § SMF for increased availability and online storage reduction § Encryption Facility for z/OS for better industry data exchange § IBM Sterling Connect: Direct® for z/OS® for better throughput and link utilization © 2017 IBM Corporation § ISV support for increased client value

IBM Offers Multiple Compression Technologies for the Mainframe Type Optimized for: Performance Overhead Supported data Frequency of access post compression CMPSC compression on IBM Z processor chip Optimal for Db 2 or select DFSMS™ files On Chip, relatively little CPU overhead and less I/O, Fast • Db 2 - Optimized for row-wise access to data is required • DFSMS files – for VSAM and non. VSAM extended format data sets Often Other software compression (zlib, or similar) Most compression uses industry std today. Used by many file types Higher CPU - software instructions executed. Note: if Java then eligible for z. IIP (or z. AAP) Any. De facto standard for almost any type of data. Often Tape HW compression Tape compression – optimized for use with large files, archival purposes Performed by the tape subsystem Any. Often – Rare (application dependent) Archival / Backup Archive data and data to backup/copy CPU overhead, longer wall clock time DFSMShsm™, DFSMSdss™ Often – Rare (application dependent) Real time compression IBM NAS storage No performance degradation SVC Designed for active primary data. z. EDC Express Active, for cross platform data exchange. Enables compression of active and inactive data Processing on z. EDC Express-expect minimal CPU overhead, low latency • • • Frequent access required. Useful for files that previously used software compression as well 16 © 2017 IBM Corporation SMF though logger zlib compatible Java BSAM/QSAM Extended format SOD DFSMShsm /dss Encryption Facility

SIMD exploitation on Linux. ONE q. Linux Kernel support v. Enable vector facility for user space applications v. Handle vector registers across context switch etc. q. Compiler & Toolchain support v. Enable application exploitation of vector facility v. Hand-optimized assembler code v. C/C++ language extension (vector types, operators, intrinsics) ü KVM 1. 1. 1 for guests exploitation ü z/VM 6. 3 + PTFs ü RHEL 7. 2 & later ü SLES 12 SP 1 & later ü Ubuntu 16. 04 & later ü Java 8 SR 1 needed o No exploitation in Java 7 Instruction pool Data pool v. Automatic vectorization by the compiler vbinutils, gcc, llvm, clang, gdb, IBM XL C/C++ compiler, Java 8 v. Optimized libraries (string, math routines) Results § IBM provided ATLAS & MASS libs, glibc Instruction pool § Application support building up Data pool Results 17 © 2017 IBM Corporation

Linux. ONE Processor Design Improvements q. Performance Enhancements q. Architecture features q. Crypto co-processor – True Random Number generator – Support for SHA 3 standard – RSA/ECC acceleration for asymmetric encryption – GCM (Galois Counter Mode) acceleration for symmetric encryption – Faster encryption and decryption • 2 X faster for most usage of AES • 6 X faster for GCM 14 nm SOI Technology Up to 10 active cores (PUs) per chip System Control (SC) chip 18 © 2017 IBM Corporation

Linux. ONE Emperor II Encryption IBM and BP Confidential – This material is for

IBM Linux. ONE Emperor II – Creating Leadership for Digital Trust q Designed with pervasive encryption for piece of mind that data and privacy is always protected q Designed with improvements in speed, efficiency and access improvements to both serve up data to build services and new offerings and to perform analytics and insight on the data because this is where it resides q Designed to be open and industry standard to bridge the skills gap and make the infrastructure easier to manage 20 © 2017 IBM Corporation IBM Linux. ONE Emperor II

Crypto Functions in the Dedicated Co-processor – new for z 14 § Dedicated § SHA-1/2 § Reduced overhead § ~2 x throughput § AES-128/256 § >4 x throughput for AES Co-Processor per Core z 196 z. EC 12 Co-Processor § Hashing hardware: ~3. 7 x better throughput (on longer data) on short data (hashing and encryption) ~1. 7 x better throughput (on longer data) z 13 § Special instructions for elliptic curve crypto / RSA z 14 § Pseudo-random number generator SHA-1/256/512, GHASH § Cipher hardware: (3)DES, AES § New hashing algorithms, e. g. SHA-3 § Support for authenticated encryption (combined encryption and hashing, e. g. AES-GCM) § True random number generator (e. g. for session keys) § New instruction added : KMGCM for end to end implementation of NIST GCM standard. (800 -38 D) § KIMD / KLMD extended to implement SHA-3 standard. (FIPS 202) § AES throughput improved to 3. 5 to 4 B/cycle 21 © 2017 IBM Corporation

Pervasive Encryption with IBM Linux. ONE Emperor II Enabled through tight platform integration Full Disk Encryption Integrated Crypto Hardware Network Encryption File System Encryption Secure Service Container 22 Full disk encryption utilizes encrypting disk drives that protect data at rest when disk drives are retired, sent for repair or repurposed Hardware accelerated encryption on every core – CPACF PCIe Hardware Security Module (HSM) & Cryptographic Coprocessor – Crypto Express Protect network traffic using standards based encryption from end to ensure that Linux systems meet approved encryption criteria Protect Linux file systems using policy controlled encryption that is transparent to applications and databases Secure deployment of software appliances including tamper protection during installation and runtime, restricted administrator access, and encryption of data and code in-flight and at-rest © 2017 IBM Corporation

Data Protection – Linux. ONE File Encryption Protection of data at-rest Value Proposition: Integration of hardware-accelerated crypto into the system’s standard components gives you wide reach into solutions DB server abc Linux. ONE block device encryption CPACF Network Storage System *** SAN Focus on Transparent Enablement: • Transparent data encryption optimized with Linux. ONE CPACF hardware performance gains • Leverage industry-unique CPACF encryption which prevents raw key material from being visible to OS and applications. dm-crypt enhancements have been submitted upstream to Linux distros for CPACF protectedkey 23 © 2017 IBM Corporation ***

Data Protection – Linux. ONE Network Security Protection of data in-flight Storage System Network *** CPACF Linux. ONE/Linux on z *** Open SSL, Java, or GSKIT abc App Client Value Proposition: Not organizations use host-based network encryption today… reduced cost of encryption enables broad use of network encryption Status: dm-crypt enhancements for CPACF protected-key submitted upstream 24 © 2017 IBM Corporation SAN Focus on Transparent Enablement: • Transparently accelerate TLS & IPSec using CPACF & SIMD to leverage hardware performance gains

Data Protection – z/VM Encrypted Paging z/VM 6. 4 *** z/VM™ xyz def CPACF SSI z/VM *** abc z/VM CPACF z/VM Single System Image cluster Value Proposition: Protect guest paging data from administrators and from users who have access to volumes 25 © 2017 IBM Corporation Encrypted Paging • Threat: access to sensitive data when stored on CP owned disk • Solution: encrypt guest data on page-out. Notes: • Paging is not SSI-relevant • Paging data does not need to survive an IPL • Ephemeral CPACF protected-key stored in CP (not on disk somewhere) • AES encryption • Very low overhead via CPACF

Crypto Express 6 S – Feature Code 0893 q One PCIe adapter per feature v Initial order – two features q Designed to be compliant with FIPS 140 -2 q q q Level 4 Installed in the PCIe I/O drawer Up to 16 features per server Prerequisite: CPACF enabling microcode (FC 3863) Support for SHA-3 Average 1. 5 X to 2 X performance increase over Crypto Express 5 S FC 0893 Three configuration options for the PCIe adapter q Only one configuration option can be chosen at any given time q Switching between configuration modes will erase all card secrets v Exception: Switching from CCA to accelerator or vice versa Accelerator CCA Coprocessor EP 11 Coprocessor* TKE N/A TKE OPTIONAL TKE REQUIRED CPACF NO CPACF REQUIRED UDX N/A UDX YES UDX NO CDU N/A CDU YES(SEG 3) CDU NO Clear Key RSA operations and SSL acceleration Secure Key crypto operations *TKE is required for EP 11 mode 26 © 2017 IBM Corporation

Data Protection – IBM Secure Service Container Extending the value of Linux. ONE hardware crypto Client Value Proposition: – Simplified, fast deployment and management of packaged solutions – Tamper protection during Appliance installation and runtime – Confidentiality of data and code running within the Appliance both at flight and at rest – Restricts administrator access to workload and data – Secure Service Container architecture builds on the value of the hardware crypto using a runtime environment designed to help clients reduce risk. Networ k 27 SAN Linux. ONE Secure Service Container © 2017 IBM Corporation Protected-key CPACF – key value not visible to OS or application

IBM Secure Service Container The Base Infrastructure to Host and Build Software Appliances q Provides simplified mechanism for fast deployment and management of packaged solutions q Provides tamper protection during appliance installation and runtime q Management provided via Remote APIs (RESTful) and web interfaces q Enables appliances to be delivered via distribution channels q Usability improvements based on customer feedback q Supported on Emperor II, Emperor, IBM z 13® GA 2, 28 © 2017 IBM Corporation Management q Ensures confidentiality of data and code running within the appliance – both at flight and at rest Services Applications Operating System

For More on IBM Secure Service Container q Presentation for Customers – IBM Secure Service Container – Aug 3, 2017 https: //w 3 -03. ibm. com/sales/support/Show. Doc. wss? docid=ZSP 04504 USEN&appname=w 3 skm q Video - Delivering highly secure solutions with IBM Secure Service Container – Jul 17, 2017 https: //w 3 -03. ibm. com/sales/support/Show. Doc. wss? docid=ZSV 03542 USEN&appname=w 3 skm q CPO Lunch and Learn - Build Hacker-proof Blockchain with IBM Secure Service Container - Apr 4, 2017 https: //yourlearning. w 3 bmix. ibm. com/#activity/URL-1490123959589 q CPO White Paper - Achieving Critical Systems Security, Governance, and Compliance with the IBM Secure Service Container – March 2017 https: //w 3 -03. sso. ibm. com/marketing/mi/compdlib. nsf/weball/81585 D 80 AECC 1 C 05002580 C 00062 F 6 F 6? Open. Document q CPO z. Spotlight: Blockchain on z leverages IBM Secure Service Containers – Mar 16, 2017 https: //w 3 -03. sso. ibm. com/marketing/mi/compdlib. nsf/weball/88 E 6 E 81357 BC 649 D 002580 BB 00675347? Open. Document 29 © 2017 IBM Corporation

Emperor II Processor Drawer Nests (2 drawers represented out of possible Models LM 1

Linux. ONE Emperor II I/O IBM and BP Confidential – This material is for

Linux. ONE Emperor II vs. Emperor CPC Drawer Structure and Interconnect Linux. ONE Emperor Fully Populated CPC Drawer Mem DIMMs Mem Mem Node 0 Node 1 CP cluster Drawer SC GX Bus 2 x PCIe PU PU GX Bus 2 x PCIe PSI X-Bus GX Bus 2 x PCIe PU PU PU GX Bus 2 x PCIe CP cluster X-Bus SC A-Bus SC PU CP cluster Drawer SC GX Bus 2 x PCIe CP cluster A-Bus CP cluster Drawer SC Mem Linux. ONE Emperor II Fully Populated CPC Drawer Mem 32 CP cluster Drawer SC 4 Drawer Linux. ONE Emperor II System Fully Interconnected ü L 1 (Instruction) increased by 33% ü L 2 (Data) increased by 100% ü L 3 increased by 100% ü L 4 reduced latency © 2017 IBM Corporation

Linux. ONE Emperor II I/O Features Supported New Build Features q. Features – PCIe I/O drawer v FICON Express 16 S+ (SX and LX, 2 SFPs, 2 CHPIDs) v OSA-Express 6 S PCIe I/O drawer • 10 Gb. E LR and SR (1 SFP, 1 CHPID) • Gb. E SX, LX, and 1000 BASE-T (2 SFPs, 1 CHPID) v 10 Gb. E Ro. CE Express 2 (new 10 Gb. E feature code) v z. EDC Express v Crypto Express 6 S 33 © 2017 IBM Corporation 32 I/O slots

A new I/O ASIC defined: An application-specific integrated circuit (ASIC), is an integrated circuit (IC) customized for a particular use, rather than intended for general-purpose use (from Wikipedia, the free encyclopedia). Used in: ‒ FICON Express 16 S+ FC/FCP/z. HPF ‒ 1 Gb. E OSA Express 6 S (LX or SX) ‒ 10 Gb. E OSA Express 6 S (LR or SR) ‒ OSA Express 6 S 1000 BASE-T ‒ 10 Gb. E Ro. CE Express 2 (new feature) Benefits ‒ ‒ ‒ 34 Up to 70% less power consumption per feature Power reduction when feature not in use Improved performance and density RAS equal or better than previous replacement features Improvement in channel latency © 2017 IBM Corporation

FICON Express 16 S+ q For FICON, z. HPF, and FCP environments § CHPID types: FC and FCP (2 PCHIDs/CHPIDs) § § 2 Gbps connectivity NOT supported FICON Express 8 S will be available to order for 2 Gbps connectivity q Auto-negotiates to 4, 8, or 16 Gbps FC 0427 – 10 KM FC 0428 – SX LX q Increased bandwidth compared to FICON Express 8 S q 10 KM LX - 9 micron single mode fiber § § Unrepeated distance - 10 kilometers (6. 2 miles) Receiving device must also be LX § § Distance variable with link data rate and fiber type Receiving device must also be SX # Start I/O per sec: 2 X-3 X OM 3 q SX - 50 or 62. 5 micron multimode fiber or q 2 channels of LX or SX (no mix) q Both ports must be FICON or FCP q 70% less Power consumption q Small form factor pluggable (SFP) optics § 35 Concurrent repair/replace action for each SFP © 2017 IBM Corporation OM 2 LX/LX OR SX/SX

Linux. ONE Emperor II I/O Connectivity Summary Features Offered As Feature Code Maximum # of features Maximum channels Increments per feature Purchase increments Storage FICON Express 16 S+ LX NB 0427 160 320 channels 2 channels FICON Express 16 S+ SX NB 0428 160 320 channels 2 channels Networking OSA-Express 6 S Gb. E LX NB 0422 48 96 ports 2 ports 1 feature OSA-Express 6 S Gb. E SX NB 0423 48 96 ports 2 ports 1 feature OSA-Express 6 S 10 Gb. E LR NB 0424 48 48 ports 1 port 1 feature OSA-Express 6 S 10 Gb. E SR NB 0425 48 48 ports 1 port 1 feature OSA-Express 6 S 1000 BASE-T NB 0426 48 96 ports 2 ports 1 feature 16 1 PCIe adapter 2 features** Crypto Express 6 S NB 0893 16 PCIe adapters Special purpose 36 10 Gb. E Ro. CE Express 2 NB 0412 8 PCIe adapters 16 ports 2 ports per adapter 1 feature z. EDC Express NB 0420 16 PCIe adapters 16 1 PCIe adapter 1 feature © 2017 IBM Corporation

Other Considerations q Linux: Linux can exploit Ro. CE Express 2 as a standard NIC (Network Interface Card) for Ethernet. A specific Linux distribution level is required (reference PSP bucket for additional details). v. SLES 11 SP 4, SLES 12 SP 2 v. RHEL 6. 8, RHEL 7. 3 v. Ubuntu 16. 04 (+ additional patches) Note. Linux does not currently support SMC-R. q Configuration or deployment issues that should be considered by field personnel or the customer In addition to the existing Ro. CE Express hardware installation procedures, when the FID is configured in HCD the Ro. CE Express 2 port number is also required. Port number must be specified. There is no default. Ro. CE Express 2 will support a greater number of Virtual Functions per physical port (63). This aspect will benefit the Linux shared Ro. CE environment. q z/VM guest support for both SMC-R and SMC-D 37 © 2017 IBM Corporation

IBM Dynamic Partition Manager © 2017 IBM Corporation

3 things to know about IBM Dynamic Partition Manager (DPM) 1. Fast v Much faster than managing with HCD and/or HCM v From hours to minutes 2. Easy v Intuitive user interface v No need for multiple administrators with different skills or tools v Do NOT expect First In Enterprise Linux clients to adopt the legacy way 3. Powerful v The same low overhead PR/SM™ hardware virtualization without the complexity v Supports dynamic configuration changes with just a few clicks of the mouse button v Provides a foundation for bare metal Cloud 39 © 2017 IBM Corporation

DPM functions via Graphical and API interfaces 40 © 2017 IBM Corporation

DPM Technical Specifications q. IBM Linux. ONE Emperor or Emperor II or Rockhopper • Feature Code #0016: Hardware Requirements for IBM DPM • Two dedicated OSA-Express 5 S 1000 BASET Ethernet q. Supported Features • FICON Express 16 S+ (Type FCP) • OSA Express 6 S • Crypto Express 6 S • z. EDC Express • Ro. CE Express 2 • Hiper. Sockets™ q. Hypervisors and Operating Systems • KVM and/or Linux on Linux. ONE • z/VM support for FCP-only configurations with z/VM 6. 4 • IBM Secure Service Container Appliances q. No support yet for • GDPS® Virtual Appliance • FICON Express 16 S+ (Type FC) q. No support for • CTC adapters are required for z/VM SSI configuration and is not supported by DPM Statement of Direction of Support of ECKD® : IBM intends to deliver support for adding and configuring ECKD FICON disks to partitions created in Dynamic Partition Manager (DPM) mode for Linux running native in an LPAR, Linux running under KVM on z, and Linux running under z/VM 6. 4. z/VM Single System Image requires FICON CTC 41 © 2017 IBM Corporation

Options for Capacity on Demand © 2017 IBM Corporation

Linux. ONE Capacity on Demand (Co. D) Capacity on Demand Permanent Upgrade –> CIU (Customer Initiated Upgrade) Temporary Upgrade Replacement Capacity Backup (CBU) Using pre-paid unassigned capacity up to the limit of the HWM No expiration Capacity - MSU % - # Engines 43 Billable Capacity (On/Off Co. D) Capacity for Planned Event (CPE) On/Off Co. D with tokens No expiration Capacity - MSU % - # Engines Tokens - MSU days - Engine days Prepaid On/Off Co. D 180 days expiration Capacity - MSU % - # Engines © 2017 IBM Corporation Post-paid On/Off Co. D with tokens 180 days expiration Capacity - MSU % - # Engines Tokens - MSU days - Engine days

Installation Planning for Linux. ONE Emperor II © 2017 IBM Corporation

Front view – Linux. ONE Emperor II Systems with 4 CPC Drawers Up to 5 I/O Drawers N+1 radiator design for Linux. ONE Air-cooled Front view – Radiator-based Air Cooled Emperor II System with 4 CPC Drawers 45 © 2017 IBM Corporation Front View – Water cooled Emperor II System with 4 CPC Drawers

Linux. ONE Emperor II Environmental vs. Linux. ONE Emperor q. Floor space – No change q. Overhead I/O or Power – No change q. Power – No change to typical power consumption q. Environment – New ASHREA A 3 classification: – 40 C°/104 F° maximum ambient temperature q. Customer Water – No change q. Weight – Slight increase in weight depending on configuration q. Airflow – No change ASHREA: American Society of Heating, Refrigerating and Air-Conditioning Engineers 46 © 2017 IBM Corporation

Available HMC configurations FC 0083 (1 U HMC) § Linux. ONE Emperor II only - Includes chassis/server/keyboard/display, maximum of 10 § Customer’s responsibility to order a rack for mounting FC 0082 (Mini. Tower) § Can be ordered as new for Linux. ONE Emperor II, maximum of 10 Note: Previous HMCs (FC 0092, 0094, 0095, and 0096) already installed can support Linux Emperor II and require appropriate driver level upgrade 47 © 2017 IBM Corporation

Operating System & Hypervisor Support Linux on z Systems IBM Linux. ONE Emperor II Minimum Distributions: • SLES 12 SP 2 • SLES 11 SP 4 • RHEL 7. 3 • RHEL 6. 9 • Ubuntu 16. 04. 1 IBM cannot legally discuss exploitation prior to GA from distributors. Officially Tested: ibm. com/systems/z/os/linux/res ources/testedplatforms. html z/VM • z/VM 6. 4 with PTFs • Optionally can include IBM WAVE for z/VM • z/VM 6. 3 with PTFs • Service withdrawn for z/VM 6. 3 on December 13, 2017 KVM for IBM z • KVM for IBM z 1. 1. 2 with latest Fixpack provides toleration mode • Can no longer order after August 28, 2017 • End of Service in March 2018 48 © 2017 IBM Corporation

The Most Highly Engineered System for Data Serving The Most Securable Linux Platform 49

Memory Configurations and Increments DDR 4 Memory DIMMS An additional 192 GB of memory is reserved above the customer purchase amount for the Hardware System Area (HSA). 51 Emperor II Model Number of CPU drawers Maximum Memory LM 1 1 8 000 GB (8 TB) LM 2 2 16 192 GB (16 TB) LM 3 3 24 384 GB (24 TB) LM 4 4 32 576 GB (32 TB) LM 5 4 32 576 GB (32 TB) q Customer offering increments are: v 64 GB memory increments from 256 GB thru 384 GB v 128 GB memory increments from 384 GB thru 896 GB v 256 GB memory increments from 896 GB thru 2944 GB v 512 GB memory increments from 2944 GB thru 32640 GB © 2017 IBM Corporation

Linux. ONE Emperor II Processor Features Drawers x IFLs Std Option Std. Model PU’s/draw IFP 2 (u. IFLs ) SAPs al SAPs Spares er LM 1 1 x 41 0 -33 (0 -32) 5 0 -4 LM 2 2 x 41 0 -69 (0 -68) 10 0 -8 LM 3 3 x 41 0 -105 (0 -104) 15 0 -12 LM 4 4 x 41 0 -141 (0 -140) 20 0 -16 LM 5 4 x 49 0 -170 (0 -169) 23 0 -16 2 1 CPs** 0– 1 ** One CP (401) is permitted only to support the GDPS Virtual Appliance, if ordered. 1. At least one IFL must be purchased in every machine and one CP is optional for use by the GDPS Virtual Appliance. 2. “u. IFL” stands for Unassigned IFL. 3. The IFP is conceptually an additional, special purpose SAP – used by PCIe I/O features. 52 © 2017 IBM Corporation

Upgrades Linux. ONE Emperor II LMx to Emperor II LMx From / To LM 2 LM 3 LM 4 LM 5 LM 1 YES YES NO LM 2 LM 3 LM 4 NO Note: Upgrading from Linux. ONE Emperor II model LM 1 -LM 4 to a model LM 5 is not supported – Factory build only. 53 © 2017 IBM Corporation

Upgrades Linux. ONE Emperor II LMx to z 14 M 0 x From /

Linux. ONE Emperor II Functions and Features (GA Driver Level 32) System, Processor, Memory I/O Subsystem, STP, Security Five hardware models (LM 1 to LM 5) Ten cores per 14 nm PU Single Chip Module (SCM) Up to 170 Integrated Facility for Linux (IFL) processors and optional System Assist Processors (SAPs) Increased Uniprocessor capacity (+10%) CPC Drawers backplane Oscillator (1 to 4 CPC drawers) PCIe Gen 3 I/O fanouts with 16 GBps Buses IBM Linux. ONE Emperor II Up to 32 TB of Redundant Array of Independent Memory (RAIM) – Up to 8 TB per CPC Drawer – 192 GB HSA GDPS Virtual Appliance Secure/Trusted SE boot and HMC New N+1 ‘radiator’ design for Air Cooled System Rack-Mounted Support Elements (SEs) in the CPC Support for ASHRAE Class A 3 datacenter New Tower or Rack-mounted HMCs and TKEs TKE 9. 0 LICC Up to 85 LPARs; up to 16 TB per LPAR (OS / hypervisor dependant) Dynamic Partition Manager (DPM) New 10 Gb. E Ro. CE Express 2 STP Enhancements RAS, Other Infrastructure Enhancements CPC Drawer/Memory Affinity and enhanced PR/SM 55 New OSA Express 6 S cards New Crypto Express 6 S with support for up to 85 Domains; and cryptographic enhancements Pause-less garbage collection for Java Enhanced processor/cache design with bigger cache sizes New FICON Epxress 16 S+ with increased start I/O rate IBM Secure Service Container for software appliances 2 nd generation SMT (for IFLs and SAPs) and enhanced SIMD Improved in-core data compression and data encryption 6 LCSS and 4 th Subchannel Set per LCSS © 2017 IBM Corporation

Memory for Linux. ONE Emperor II Model LM 1 (1 CPC Drawer) 56 Feature Customer Purchase 1660 256 GB 1661 320 GB 1662 384 GB 1663 448 GB 1664 512 GB 1665 576 GB 1666 704 GB 1667 832 GB 1668 960 GB 1669 1088 GB 1670 1216 GB 1671 1344 GB 1672 1472 GB 1673 1600 GB 1674 1856 GB 1675 2112 GB 1676 2368 GB Feature Customer Purchase 1677 2624 GB 1678 2880 GB 1679 3136 GB 1680 3392 GB 1681 3648 GB 1682 3904 GB 1683 4116 GB 1684 4928 GB 1685 5440 GB 1686 3962 GB 1687 6464 GB 1600 GB 1688 6076 GB 1856 GB 1689 7488 GB 1690 8000 GB Dial Max 320 GB 576 GB 832 GB 1088 GB 1344 GB 2368 GB © 2017 IBM Corporation Dial Max 2880 GB 3392 GB 3904 GB 4928 GB 3952 GB 6976 GB 8000 GB

Memory for Linux. ONE Emperor II Model LM 2 (2 CPC Drawers) 57 Feature Customer Purchase Dial Max Feature Customer Purchase 256 GB 1677 2624 GB 1695 10560 GB 1661 320 GB 1678 2880 GB 1662 384 GB 1679 3136 GB 1696 11072 GB 1663 448 GB 1680 3392 GB 1697 11584 GB 1664 512 GB 1681 3648 GB 1698 12096 GB 1665 576 GB 1682 3904 GB 1666 704 GB 1683 4416 GB 1699 12608 GB 1667 832 GB 1684 4928 GB 1700 13120 GB 1668 960 GB 1685 5440 GB 1686 5952 GB 1701 13632 GB 1687 6464 GB 1702 14144 GB 1688 6976 GB 1703 14656 GB 1689 7488 GB 1690 8000 GB 1704 15168 GB 1691 8512 GB 1705 15680 GB 1692 9024 GB 1693 9536 GB 1706 16192 GB 1694 10048 GB Feature Customer Purchase 1660 1669 1088 GB 1670 1216 GB 1671 1344 GB 1672 1472 GB 1673 1600 GB 1674 1856 GB Dial Max 832 GB 1088 GB 1344 GB 1600 GB 1856 GB 1675 2112 GB 1676 2368 GB 9024 GB 10048 GB © 2017 IBM Corporation Dial Max 11072 GB 12096 GB 13120 GB 14144 GB 15168 GB 16192 GB

Memory for Linux. ONE Emperor II Model LM 3 (3 CPC Drawers) Feature Customer Purchase Dial Max Feature Customer Purchase 256 GB 1678 2880 GB 1696 11072 GB 1713 19776 GB 1661 320 GB 1679 3136 GB 1697 11584 GB 1662 384 GB 1680 3392 GB 1698 12096 GB 1714 20288 GB 1663 448 GB 1681 3648 GB 1699 12608 GB 1664 512 GB 1682 3904 GB 1700 13120 GB 1715 20800 GB 1665 576 GB 1683 4416 GB 1701 13632 GB 1716 21312 GB 1666 704 GB 1684 4928 GB 1702 14144 GB 1667 832 GB 1685 5440 GB 1703 14656 GB 1717 21824 GB 1668 960 GB 1686 5952 GB 1704 15168 GB 1669 1088 GB 1687 6464 GB 1705 15680 GB 1718 22336 GB 1670 1216 GB 1688 6976 GB 1706 16192 GB 1719 22848 GB 1671 1344 GB 1689 7488 GB 1707 16704 GB 1672 1472 GB 1690 8000 GB 1708 17216 GB 1720 23360 GB 1673 1600 GB 1691 8512 GB 1709 17728 GB 1674 1856 GB 1692 9024 GB 1710 18240 GB 1721 23872 GB 1675 2112 GB 1693 9536 GB 1711 18752 GB 1722 24384 GB 1676 2368 GB 1694 10048 GB 1712 19264 GB 1677 2624 GB 1695 10560 GB Feature Customer Purchase 1660 58 Dial Max 1344 GB 1600 GB © 2017 IBM Corporation 13120 GB 14144 GB 15168 GB 16192 GB 17216 GB 18240 GB 19264 GB Dial Max 20288 GB 21312 GB 22336 GB 23360 GB 24384 GB

Memory for Linux. ONE Emperor II Model LM 4/LM 5 (4 CPC Drawers) (Pt 1) 59 Feature Customer Purchase Dial Max 1660 256 GB 1678 2880 GB 1661 320 GB 1679 3136 GB 1662 384 GB 1680 3392 GB 1663 448 GB 1681 3648 GB 1664 512 GB 1682 3904 GB 1665 576 GB 1683 4416 GB 1666 704 GB 1684 4928 GB 1667 832 GB 1685 5440 GB 1668 960 GB 1686 5952 GB 1669 1088 GB 1687 6464 GB 1670 1216 GB 1688 6976 GB 1671 1344 GB 1689 7488 GB 1672 1472 GB 1690 8000 GB 1673 1600 GB 1691 8512 GB 1674 1856 GB 1692 9024 GB 1675 2112 GB 1693 9536 GB 1676 2368 GB 1694 10048 GB 1677 2624 GB 1695 10560 GB Dial Max 1856 GB © 2017 IBM Corporation

Memory for Linux. ONE Emperor II Model LM 4/LM 5 (4 CPC Drawers) (Pt 2) 60 Feature Customer Purchase Dial Max Feature Customer Purchase 1696 11072 GB 1713 19776 GB 1697 11584 GB 1714 20288 GB 1698 12096 GB 1715 20800 GB 1699 12608 GB 1716 21312 GB 1700 13120 GB 1717 21824 GB 1701 13632 GB 1718 22336 GB 1702 14144 GB 1719 22848 GB 1703 14656 GB 1720 23360 GB 1704 15168 GB 1721 23872 GB 1705 15680 GB 1722 24384 GB 1706 16192 GB 1723 24896 GB 1707 16704 GB 1724 25408 GB 1708 17216 GB 1709 17728 GB 1725 25920 GB 1710 18240 GB 1726 26432 GB 1711 18752 GB 1727 26944 GB 1712 19264 GB 1728 27456 GB 17216 GB 18240 GB 19264 GB Feature Customer Purchase 1729 27968 GB 1730 28480 GB 1731 28992 GB 22336 GB 1732 29504 GB 23360 GB 1733 30016 GB 1734 30528 GB 1735 31040 GB 1736 31552 GB 1737 32064 GB 1738 32576 GB Dial Max 20288 GB 21312 GB 24384 GB 25408 GB 26432 GB 27456 GB © 2017 IBM Corporation Dial Max 28480 GB 29504 GB 30528 GB 31552 GB 32576 GB

Hardware Crypto support in IBM Linux. ONE Architecture Linux. ONE CPC Drawers Linux. ONE PU SCM CPACF Crypto Express 6 S PCIe I/O drawers Rack Mounted or Tower Trusted Key Entry 61 IBM Linux. ONE Emperor II Smart Cards Smart Card Readers © 2017 IBM Corporation

Five Linux. ONE Emperor II models Machine Type 3906 Model Air Water # PU’s LM 1 FC 1107 FC 1112 33 LM 2 FC 1108 FC 1113 69 LM 3 FC 1109 FC 1114 105 LM 4 FC 1110 FC 1115 141 LM 5 FC 1111 FC 1116 170 Air = Radiator cooled Water = Water cooled 62 © 2017 IBM Corporation

IBM Linux. ONE Availability Dates q The following are available as of September 13, 2017 v IBM Emperor II Models LM 1, LM 2, LM 3, LM 4, and LM 5 v Field installed features and conversions on IBM Emperor II that are delivered solely through a modification to the machine's Licensed Internal Code (LIC) v TKE 9. 0 LIC (#0879) on IBM Linux. ONE Emperor II, IBM Linux. ONE Emperor™, and IBM Linux. ONE Rockhopper ™ v TKE Rack Mount w/4768 (#0085) on Emperor II, Emperor, and Rockhopper v TKE w/4768 (#0086) on Emperor II, Emperor, and Rockhopper v HMC (#0082) on Emperor II, Emperor, and Rockhopper v HMC Rack Mount (#0083) on Emperor II, Emperor, and Rockhopper 63 © 2017 IBM Corporation Linux. ONE Emperor II

IBM Linux. ONE Availability Dates q Available December 15, 2017 v z/VM Guest exploitation support for the Instruction Execution Protection Facility v z/VM Guest exploitation support for pause-less garbage collection v z/VM support for encrypted paging q Available December 31, 2017 v MES features for IBM Emperor II Models LM 1, LM 2, LM 3, LM 4, and LM 5 v MES upgrades for IBM Emperor II LMx models to IBM z 14 M 0 x models Linux. ONE Emperor II 64 © 2017 IBM Corporation

IBM Linux. ONE Emperor II: Designed for Trusted Digital Experiences Pervasive Encryption is the New Standard • 100% encryption of all data • Zero application changes • Zero impact to service levels • Protect client and corporate from internal and external threats Analytics & Machine Learning for Continuous Intelligence across the Enterprise • Anticipate customer needs and embed insight in every business transaction • Dramatically faster lifecycle management of behavioral models with more memory and greater processing capacity • Derive impactful insights by combining IBM Linux. ONE data with other structured and unstructured external data sources 65 © 2017 IBM Corporation Open Enterprise Cloud to Extend, Connect and Innovate • Cut new service build time by 90% using secure APIs on IBM Linux. ONE and advanced Dev. Ops • Seamlessly connect any service from public and private cloud with transactions and data on IBM Linux. ONE • Accelerate innovation with an ecosystem of partners to develop and manage enterprise wide applications leveraging 1000’s open source software packages

Quick View of IBM Linux. ONE Emperor II q New cover, no vertical stripe, Linux. ONE badge relocated q Five models: LM 1, LM 2, LM 3, LM 4 and LM 5 q Processor Units (PUs) v v v v 5. 2 GHz 41 (49 for LM 5) PU cores per CPC drawer Single thread capacity ~1832 PCI 33, 69, 105, 141 or 170 IFL processors available for Linux workloads Up to 23 SAPs per system, standard New on Linux. ONE Emperor II is SAPs are SMT Enhanced performance for compression and crypto coprocessor Unchanged from Emperor I § 2 spares designated per system § 1 CP available (option) to support the GDPS Virtual Appliance § Up to 85 LPARs q Memory v Redundant Array of Independent Memory, like Emperor I v System Min of 256 GB - up to 8 TB / drawer - 192 GB Fixed HSA, standard v Up to 32 TB for System and up to 16 TB per LPAR (OS dependent) q I/O v v 66 Now a single System Control Chip serves each drawer Technology refresh of all I/O cards (FICON, OSA, Ro. CE) 16 GBps PCIe Gen 3 I/O Interconnects 6 Logical Channel Subsystems (LCSSs) with 4 Sub-channel sets per LCSS © 2017 IBM Corporation IBM Linuc. ONE Emperor II Machine Type: 3906 Models: LM 1, LM 2, LM 3, LM 4, Max IFLs Model LM 5 Memory LM 5 170 32 TB LM 4 141 32 TB LM 3 105 24 TB LM 2 69 16 TB LM 1 33 8 TB*

Cross-Sell: IBM Linux. ONE with IBM Storage and IBM Flash. System IBM Storage Solution Support for Linux. ONE § KVM on z, IBM z/VM, Linux on z (= “bare metal”, native, LPAR) IBM Flash. System A 9000/A 9000 R § FCP connectivity only § z. High Performance FICON (z. HPF) and Extended Address Volumes § GDPS/PPRC Hyper. Swap® ® IBM System Storage DS 8880 § FICON Express 16 S with Forward Error Correction Codes, FICON Dynamic Routing, Read Diagnostic Parameters, Enhanced Write Protocol XIV® Storage Systems § IBM z/VM, Linux on z - FCP connectivity only Storwize® V 7000 § KVM on z, z/VM, and Linux on z - FCP connectivity only SVC § KVM on z, z/VM, and Linux on z - FCP connectivity only IBM Flash. System V 9000 § KVM on z, z/VM, and Linux on z - FCP connectivity only IBM Flash. System 900 § KVM on z, z/VM, and Linux on z - FCP connectivity only § z/VM required for its channel protocols; KVM not supported, native Linux not supported TS 7700 § FICON connectivity only § FCP can attach to Linux on z GDPS support for Hyper. Swap via the GDPS Virtual Appliance for Linux running under z/VM or KVM (GDPS appliance not available for native Linux on z) § Support for Linux on z, and z/VM, via FICON if TS 7700 LUP 12387 -USEN-00 is front 67 TS 1150 , TS 4500 © 2017 IBM Corporation 1 67