Global Workload Manager Overview

Size: px
Start display at page:

Download "Global Workload Manager Overview"

Transcription

1 Global Workload Manager Overview Dan Herington Infrastructure Solutions Division 2004 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice

2 The Adaptive Enterprise Business and IT synchronized to capitalize on change Business Information Technology Business benefits: simplicity, agility, value

3 Why You Need an Adaptive Infrastructure Tremendou s amount of unutilized capacity Yet these systems are unable to handle the load 0 Server Average Utilization Most reports put average utilization at approximately 30% Some applications still not able to meet performance requirements

4 Why Utilization is So Low Mission Critical applications have peaks in load that require spare capacity

5 New Alternatives to Overprovisioning Build an Adaptive Infrastructure that can dynamically react to application peaks Implement Application Consolidation Run many workloads on a small number of servers Share Spare Capacity Provide spare capacity for multiple apps on the same system or systems 40% reduction in spare capacity requirements Consolidation requires virtualization solutions to ensure applications are isolated from each other

6 Adaptive Enterprise vision Business and IT synchronized to capitalize on change 1. Measure, assess and maintain a dynamic link between business and IT 2. Architect and integrate heterogeneous IT environments 3. Extend and link business processes across suppliers and customers 4. Manage and control business processes, applications and the whole IT environment MEASURE & ASSESS time, range, ease Business processes EXTEND & LINK Suppliers Employees Customers Applications ARCHITECT & INTEGRATE simplify, standardize, modularize, integrate Virtualized resources Infrastructure MANAGE & CONTROL assess, advise, act

7 HP s Definition of Virtualization An approach to IT that pools and shares resources so utilization is optimized and supply automatically meets demand Business Supply Demand Information technology

8 gwlm Overview

9 gwlm Vision gwlm enables utility computing by supporting centralized resource-sharing policies (definition, enforcement and monitoring) on HP servers. The gwlm Design Center Across different partitioning and virtualization approaches Across multiple systems datacenter wide Supporting different operating systems - HP- UX, Linux in 1.0, Windows and OpenVMS (future) For scale up and scale out utilities 8 Single-cell NPars Superdome 1 Superdomes 2 thru 6 8 Cells, 32 CPU, 64 Gig 8 Cells, 32 CPU, 64 Gig For fault isolation, we try to NPar NPar NPar NPar only run 1 production workload per NPar. Each NPar: 1 Cell 4 CPU NPar NPar NPar NPar 8 Gig Mem 2 s NPar 1 Gig Gig Enet CPU 1 Gig Gig Enet NPar NPar NPar CPU 1 Gig FC (to SAN) 1 Gig FC (to SAN) 1 Gig Gig Enet CPU 1 Gig Gig Enet CPU 1 Gig FC (to SAN) 1 Gig FC (to SAN) NPar NPar NPar Typical : 2 CPU 4 Gig Mem 2 Gig enet What s running: 2 FC Lots of 3-tier apps. The databse for these is usually very small. Lots of Oracle, SAS, WebLogic Server. Some Ixos, some SAP 4 Single-cell NPars (2 s each) 2 dual-cell NPars (3 s currently) 1 Gig 1 Gig Gig Enet CPU 1 Gig 1 Gig Gig Enet CPU 1 Gig 1 Gig FC (to SAN) 1 Gig 1 Gig FC (to SAN) Oracle s : 2 CPUs, 8 Gig Mem 20 Oracle Instances 350 Meg / inst. Simplicity through pre-defined policies, central management, ease of use Increased agility by dynamically allocating server resources on-the-fly to meet customer's changing demands Delivering additional value by enabling more efficient utilization of shared servers

10 gwlm 1.0 Features and Functions Improving CPU utilization dynamic, policy-based CPU allocation Manage s, psets, and FSS groups on HP-UX 11i v1 for HP 9000 servers 11i v2 for HP Integrity and HP 9000 servers Manage psets on Linux 2.6 Kernels HP Integrity Servers only at first release

11 gwlm 1.0 Features and Functions Ease of management for large numbers of systems central management server web-based GUI integrated with HP Systems Insight Manager central policy management central monitoring and reporting real time and historical utilization, demand, & resource allocation discovery of virtual partitions and resource partitions out of box support for common, easy to understand configurations own/borrow policy fixed entitlement policy CPU utilization policy single policy can be applied to multiple resource partitions most sites only need several variations of factory supplied policies role-based access Administrator mode to create/apply policies and configurations Monitor only mode for viewing real-time or historical workload statistics Goal: customer can manage s environment in under 30 minutes after software install Customer test showed customer could actively manage environment in 15 minutes without quick start guide customer test result: customer who read quick start guide was managing workloads in 5 minutes

12 gwlm 1.0 Features and Functions Building Trust in dynamic capabilities ensure tool is configured as planned and will behave as expected Advisory mode Realtime Reporting Historical reports application owners can see results CPU resources can be guaranteed Fixed Policy Own/Borrow (PolicyMin, owned) Utilization (PolicyMin) Custom (Goal Mode) Audit reports to prove each workload got what was guaranteed show when workloads got more resources than the guaranteed minimum Scheduled for August 04 announcement and December 04 shipment Customer Validation and Early Access Programs available now

13 gwlm Policies OwnBorrow Specify minimum, maximum and owned resources gwlm will ensure there is enough resource to satisfy all workloads at the owned level Workloads can share unused resources, but are guaranteed access to the owned amount if needed Fixed Specify a fixed amount of resource for this workload No sharing will be allowed Utilization Specify minimum, maximum and a utilization target If actual utilization is higher than target, policy will request more, if below target policy will request less Custom (Goal Mode) Specify a min, max and metric (e.g. response time) for goal mode allocation of CPU

14 gwlm Out of the Box Reports Troubleshoot a poorly performing workload Get periodic capacity and performance report Produce a resource audit report for internal customers Police my internal customers identify resource hogs Right-size a workload s entitlement

15 Integrated with HP Systems Insight Manager Task oriented graphical user interface Each shared environment is a Shared Resource Domain Integrated event system to track SLO violations Configuration panel, monitoring panels also available Factory-defined policies simplify configuration of shared domains Multiple shared resource domains can be managed from the central management station Abstraction of different types of containers, vpars, psets, Auto-discovery of containers

16 gwlm utilization monitoring Selection screen for monitoring capabilities with gwlm Graphs with performance against policy Real-time and historical utilization graphs for each workload

17 gwlm Architecture

18 HP Global Workload Manager Concepts Server resources are organized into Shared Resource Domains (SRDs) which can be managed in either active or advisory mode Workloads are deployed into flexible containers (resource partitions or s in release 1.0) within the SRDs Policies are defined centrally and can be applied to workloads in multiple SRDs Demand and supply of resources to each workload is regulated by the policies in each SRD Monitoring and management of the utility is accessed from a central management console providing breakthrough ease of use

19 gwlm High Level Architecture Central Management Station (CMS) Historica l And RT Data HP Systems Insight Manager gwlm Console Policy Config Data gwlm Node Mgr gwlm Node Mgr gwlm Node Mgr gwlm Node Mgr gwlm Node Mgr gwlm Node Mgr Linux PSETs npar or system with s npar w/rps npar w/rps Managed Nodes

20 gwlm Central Management Station Web-based GUI integrated with Systems Insight Manager Container Discovery Workload Definition Policy Configuration Policy Deployment Workload Monitoring Real-Time and Historical Reports CMS Components Web-based GUI Services CMS Daemon Stores data and provides startup services for node managers Data Repository Discovery, policies and utilization data

21 gwlm Node Manager Discovery Discovery of containers and workloads in this OS image Workload Controllers Collect data about each workload (status, metrics, utilization) Calculate resource requirements to satisfy policies Shared Resource Domain Management and Arbitration Collect data for all workloads in this OS image Communicate local workload data to Master for inter-os SRDs Arbitrate and reallocate resources for sub-os SRDs Process Management Moves processes to the appropriate sub-os container for their workload Data Collection and Aggregation Utilization and Metric data is collected for all workloads on the OS image and is passed up to the gwlm daemon on the CMS for storage in the database Alarms are passed up to the gwlm daemon and integrated with other alarms in the Systems Insight Manager alarm system

22 gwlm Graphical User Interface

23 gwlm Policy Monitoring Graph

24 gwlm Policy Constraint Graph

25 gwlm Workload Container Utilization Graph

26 gwlm Configuration Wizard

27 gwlm Menu Screenshot

28 gwlm Wiz Enter Systems Screenshot

29 gwlm Wiz Select Container Screenshot

30 gwlm Wiz Review SRD Screenshot

31 gwlm Wiz Policy Screenshot

32 gwlm Wiz Review Screenshot

33 gwlm Wiz Next Steps Screenshot

34 Where to get gwlm

35 Obtaining gwlm gwlm 1.0 will be available for download from software.hp.com in December, 2004 It will also be shipped on the March 2005 Application Release install media gwlm will be licensed on a per-processor basis for the managed nodes only investigating site license option

36 How Does this Compare to HP-UX WLM HP-UX WLM will continue to be offered and will continue to be part of the HP-UX Mission Critical OE. When to purchase gwlm When IT serves as a service provider to BU s Whenever you are trying to manage a large infrastructure with many similarly configured systems or workloads When the customer wants support for Linux PSETs on Integrity Servers When to purchase WLM When the BU owns the systems and wants to control resource sharing When there is little commonality in the way resources are managed and shared between different managed systems When there is a small number of systems being managed When PPU or TiCOD management is required When npar/icod management is required

37 gwlm Demo

38 The Demo Scenario IT manages a server utility with 20 servers each set up with 3 s They offer the infrastructure as a managed service to their business units They want to increase utilization so they are allowing the sharing of unused resources by the workloads on each server Each workload owns a certain number of CPUs and that is how the business units are charged for the service

39 The gwlm Demo Solution 3 -based Workloads on an 8 CPU server A Production Sales Application Owns 3 CPUs Max 6 A Production Finance Application Owns 4 CPUs Max 6 A Dev/Test Environment Owns 1 CPU Max 6 We will show 3 demos: Creating the policies required Configuring gwlm to manage these workloads Monitoring gwlm s allocation of CPUs as the loads on the applications varies over time

40 Co-produced by: