Partition Load Manager

IBM System p5

IBM Power Systems p AIX/UNIX/Linux Virtualization Basics

Key Terminology: Practical Implementations of Virtualization & Live Partition Mobility

1 IBM System p5 Server Workload Considerations

Traditional Server Purchase Capacity 6 –Workloads fluctuate 5 differently throughout the Peak day and year 4 –Workloads can fluctuate 3 greatly moment to CPUs moment 2

–Servers are purchased to 1 handle individual unknown peaks 0 –Unused resources can not be moved between 7:00 9:00 11:00 1:00 3:00 5:00 servers Time

2 LPAR (LogicalIBM System p5Partitions) – Separate O/S Images AIX Workload Manager (WLM) Multiple Applications (processes) controlled (fenced off) with a single AIX image Improved administrative efficiency by reducing the number of AIX images to maintain

CPU

Memory

CPU

Memory

3 IBM System p5 AIX Workload Partitions Separate regions of application space within a single AIX image Improved administrative efficiency by reducing the number of AIX images to maintain

Workload Software partitioned system capacity Partition –Each Workload Partition obtains a regulated share of Test system resources Workload Workload –Each Workload Partition can have unique network, Partition Partition filesystems and security Billing Application Two types of Workload Partitions Server –System Partitions Workload –Application Partitions Partition Web Separate administrative control Server Workload Workload Partition –Each System Workload partition is a separate Partition Test administrative and security domain BI Shared system resources Operating System, I/O, Processor, Memory AIX 6.1

4 IBM System p5 PowerVM AIX Virtualization Continuum

AIX Workload Partitions Complement Logical Partitions LPAR / Micropartitions AIX V5.3 on POWER5 or later

Workload Partitions AIX 6 on POWER4 or later

AIX Workload Manager Resource Flexibility AIXV4.3.36 on POWER3 or later

Workload Isolation

5 IBM System p5 Two WPAR AIX Offerings…

AIX 6 – Workload Partitions (WPAR) included in AIX 6 – Element (single system) WPAR Management

Workload Partitions Manager™ – Enablement for Live Application Mobility – Cross System Management for Workload Partitions – Automated, Policy-based Application Mobility – Part of IBM System Director Family

WPAR Manager

6 IBM System p5 Workload Partitions Manager

Browser

Management of WPARS across multiple systems Lifecycle operations Single Console for: – Graphical Interface – Create & Remove Workload – Start & Stop Partition – Checkpoint & Restart Manager – Monitoring & Reporting – Manual Relocation – Automated Relocation – Policy Driven Change Infrastructure Optimization Web Service Load Balancing Server 1 Server 2 Server 3 WPAR Agent WPAR Agent WPAR Agent

System/Application WPARs System/Application WPARs System/Application WPARs

7 IBM System p5 Graphical WPAR Manager & Application Mobility

Workload Partition Manager

8 IBM System p5 AIX Live Application Mobility Move a running Workload Partition from one server to another for outage avoidance and multi-system workload balancing

Workload Workload Partition Workload Partition App Server Partition Data Mining e-mail Workload Workload Partition Partition Workload QA Web Partition Workload Dev Partition Billing Workload AIX Partitions AIX Manager Policy

Works on any hardware supported by AIX 6, including POWER5 and POWER4

9 IBM System p5 IBM System Simultaneous multithreading

POWER5POWER4 & POWER6 (Single Threaded) (simultaneous multithreading)

FX0 Thread0 active FX1 No thread active LSO Thread1 active LS1 FP0 Appears as four CPUs per chip to the FP1 operating system BRZ (AIX 5L V5.3 and System throughput ST CRL Linux) SMT

Utilizes unused execution unit cycles Presents symmetric multiprocessing (SMP) programming model to software Natural fit with superscalar out-of-order execution core Dispatch two threads per processor: “It’s like doubling the number of processors.” Net result: – Better performance – Better processor utilization

10 IBM System p5

IBM partitioning innovations help solve this problem

In 2001: Logical Partitioning enabled consolidation of multiple application workloads

In 2002: Dynamic Logical Partitioning, enabled dynamic reassignment of workloads

Sys3 LogicalLogical DynamicDynamic

Application A Application B Application C Sys2

Sys1

11 IBM System p5 Meeting Demand: Utilizing Available CPU Cycles

Policy-based, automatic partition resource tuning Dynamically adjust CPU and memory allocation

Before resource tuning After resource tuning Unbalanced resource allocation Adjust resource allocation based on business priority

3 5 6 1 3 10 CPUs CPUs CPUs CPU CPUs CPUs

Agent Agent Agent Agent Agent Agent

Test LPAR CRM LPAR Finance LPAR Test LPAR CRM LPAR Finance LPAR

12 IBM System p5

Micro-Partitioning technology allows Micro-Partitioning technology each processor to be subdivided into as AIX 5.3/AIX 6/Linux/i5 & iOS many as 10 “virtual servers”, helping to Micro-partitions consolidate UNIX® and Linux Dynamic applications. LPARs Pool of 6 CPUs Whole Partitioning options Processors – Micro-partitions: Up to 254* – Dynamic LPARs: Up to 32* – Combination of both Linux Linux Configured via the HMC AIX 5L V5.3 AIX 5L V5.3 AIX 5L V5.2 AIX 5L V5.3 i5/OS V5R3** Number of logical processors – Minimum/maximum

Entitled Entitled capacity capacity – In units of 1/100 of a CPU – Minimum 1/10 of a CPU Min Variable weight Max – % share (priority) of surplus capacity Hypervisor Capped or uncapped partitions *on p5-590 and p5-595 ** on p5-570, p5-590, and p5-595 13 IBM System p5 More Efficiency: CPU & I/O Sharing

Dynamically resizable Included features: 3 5 6 6 Enablement for Micro-Partitioning CPUs CPUs CPUs CPUs – Share processors across multiple partitions Virtual I/O Micro-partitioning th server – Minimum partition 1/10 processor AIX 5L AIX 5L – AIX 5L V5.3 or Linux* partition AIX 5.2 V5.2 V5.3 Virtual I/O server Storage – Shared Ethernet sharing Linux Linux AIX 5L V5.3 AIX 5L V5.3 AIX 5L V5.3 – Shared SCSI and AIX 5L V 5.3 Ethernet Fibre Channel-attached disk subsystems sharing Server I/O Virtual – Supports AIX 5L V5.3 and Linux* Virtual I/O Partition Load Manager paths – Both AIX 5L V5.2 and Hypervisor AIX 5L V5.3 supported – Balances processor and memory request Partition Load Manager Unmanaged partitions partitions

LPAR 1 LPAR 2 LPAR 3 Hardware Management PLM AIX 5L V5.2 AIX 5L V5.3 Linux Console Management PLM agent PLM agent (HMC) Advanced server POWER Hypervisor Virtualization * SLES 9 or RHEL AS 3 hardware feature

Jay Kruemcke IBM 2003 IBM System p5 POWER 5 & 6 / AIX 5.3 LPAR DISK Sharing

Partition Partition Partition Partition Partition 1 2 3 4 5 I/O Server Hypervisor Layer I/O Server 1A 2A 3A 4A 5A

2A 2B SCSI 3A 3B SCSI A 4A 4B B Fibre 1B 2B 3B 4B 5B Fibre Hypervisor Layer Mirror Mirror Mirror Mirror Mirror

A A A A A A A A A A

Using LVM mirroring for Availability One physical drives appears to be multiple logical drives

LUNs appears as individuals logical drives Minimizes the numbers adapters SCSI and Fibre supported

15 IBM System p5 POWER5 / AIX 5.3 LPAR Virtual Ethernet

Partition Partition Partition Partition Partition 1 2 3 4 5

TCP/IP TCP/IP TCP/IP TCP/IP TCP/IP Stack Stack Stack Stack Stack I/O Server I/O Server Partition Partition VLAN

Power VLAN VLAN Hypervisor R IP B Network

Configured like a standard Ethernet

IP forwarding provided by I/O Server partition

Can have multiple connections per partition

Inter-LPAR communications: IPv4 & IPv6 plus large MTUs

16 IBM System p5

Virtual Ethernet helps reduce hardware costs by Virtual networking sharing LAN adapters POWER5 Server External Servers Micro-partitions VIOS AIX 5L AIX 5L Linux Linux VLAN 1 vSCSI V5.3 V5.3 VLAN 2 VLAN 3 vLAN

Shared Ethernet Adapter Ethernet Switch Virtual Ethernet Switch

Virtual Ethernet POWER Hypervisor –Partition to partition communication –Requires AIX 5L V5.3 and POWER5 Shared Ethernet Adapter IEEE VLANS –Provides access to outside world –Up to 4096 VLANS –Uses Physical Adapter in the Virtual I/O Server –Up to 65533 vENET adapters VLAN – Virtual LAN –21 VLANS per vENET adapter –Provide ability for one adapter to be on multiple subnets –Provide isolation of communication to VLAN members –Allows a single adapter to support multiple subnets Available via optional Advance POWER Virtualization or POWER Hypervisor and VIOS features.

17 IBM System p5 Virtual SCSI Virtual I/O helps reduce hardware costs by sharing disk drives POWER5 Server External Storage Micro-partitions Shared VIOS AIX 5L Linux AIX 5L Linux Fiber Chan v v V5.3 V5.3 A1 A2 A3 A4 Adapter S C L A5 A A1 A2 Shared S N B1 B2 B3 A3 SCSI I Adapter B1 B2 B4 B5 B3 Virtual SCSI VIOS owns physical disk resources LVM based storage on VIO Server POWER Hypervisor Physical Storage can be SCSI or FC Local or remote

Micro-partition sees disks as vSCSI (Virtual Multiple LPARs can use same or different physical SCSI) devices disk Virtual SCSI devices added to partition Configure as logical volume on VIOS via HMC Appear a hdisk on the micro-partition LUNs on VIOS accessed as vSCSI disk Can assign entire hdisk to a single client VIOS must be active for client to boot Available via optional Advance POWER Virtualization or POWER Hypervisor and VIOS features.

18 IBM System p5

On Demand Capabilities: Keeps on Expanding

CoD Offerings (Capacity on Demand) Trial Reserve On/Off Backup RAS (Reliability Availability Serviceability) Processor and Memory Sparing Building Block Architecture Pay as you grow

19 IBM System p5 Capacity on Demand Capacity Upgrade on Demand Upgrade system with processors and/or memory No special contracts, no required monitoring (no ability to turn off the capacity) Purchase agreement On/Off Capacity on Demand Temporary use of requested number of processors or amount of memory Client selects the capacity and activates the resource (registered system) Capacity can be turned on and off by the client Information captured by IBM (or reported to IBM) Rental agreement Reserve Capacity on Demand Processor resources only (processor days) Capacity can be turned on and off by the client Prepaid debit agreement Requires AIX 5L V5.3 and APV Trial Capacity on Demand Allow clients to test the effects of additional processors and/or memory Partial or total activation of processors and memory Resources available for fixed time No formal commitment required Dynamic Processor Sparing • Automated replacement of de-allocated processors • Unassigned or inactive processors

20 IBM System p5

First Failure Data Capture Competitive Strategy - Failure Re-create Error Checkers ƒ Run diagnostic testcases during service call ƒ Repair based on testcase symptom Fault Isolation Register (FIR) CPU ƒ Questionable correlation to original customer problem L1 Cache ƒ Development focus on improved testcases ƒ Open service action plan if failure not re-created L2 Cache Service Processor ƒ Testcases used to isolate failures and verify Log Error correct operation Nonvolatile First Failure Data Capture Memory RAM ƒ Specialized hardware designed to capture failure data at the time of failure ƒ Repair based on root-cause analysis ƒ Direct correlation to original problem Disk ƒ Engineering focus on built-in error detection and capture ƒ Service action plan driven by captured failure Reliably identify failing component information reducing costly downtime ƒ Testcases used only to verify operation

21 IBM System p5 POWER RAS additions…

Primary POWER 2005 RAS Enhancements RAS features

Service Processor Failover: 9 First Failure Data Capture – p5-570, p5-590 and p5-595 9 DDR Chipkill™ memory – New service processor option for p5-570 9 Bit-steering/redundant memory 9 Required for “Failover” support 9 Memory soft scrubbing 9 Redundant power, fans Dynamic Firmware Maintenance: 9 Dynamic Processor Deallocation – Apply firmware fixes without system disruption 9 Dynamic processor sparing – Fixes only, not new functionality 9 ECC memory 9 Persistent memory deallocation Hot I/O Drawer Add Hot-plug PCI slots, fans, power 9 – Install remote I/O drawers without system 9 Internal light path diagnostics disruption 9 Hot-swappable disk bays 9 I/O error handling extended beyond base PCI adapter 9 ECC extended to inter-chip connections for the fabric/processor buses HMC required to enable these functions..

22 IBM System p5 Enable high system availability – Building on world-class hardware RAS

Summary of key pSeries RAS features

Core System Design Fault Resilience System Restore yHigh quality parts yN+1 Power Supplies, yDeferred Repair yFewer parts = Fewer regulators, power cords yConcurrent Repair failures yDual redundant fans yLED Service Identification yDesigned for low power yDynamic Processor yService Consoles consumption (less heat = Deallocation and yMigration to Guided fewer failures) sparing Maintenance yManufacturing methods, y"Chipkill" Technology packaging, cooling yPredictive Failure Fault Isolation & Diagnosis yContinuous System and Analysis yFirst Failure Data Capture Commodity Quality yAuto Path yRun Time Self Diagnostics Actions Reassignment - data yService Processor yIntegrated RAS features paths, power yRifle-shot repairs (no "plug yFailure Avoidance and pray" parts replacement Methodology approach) yDesigned for Ease of Service

23 IBM System p5

Summary

Why IBM System p AIX/LINUX solutions

–IBM has a Proven Technology –IBM has a Consistent Product line –IBM has a Clear Direction –A good hockey player plays where the puck is. A great hockey player plays where the puck is going to be. –Wayne Gretzky

–IBM wants to EARN your business