1 Autonomic Computing IBM Initiatives in Autonomic Computing

Today’s Complex Infrastructure Business Data UNIX Mainframe Web Servers PCs SSL Appliances Routers Switches Firewall Servers Security & Directory Servers Application Servers UNIX PCs Caching Appliances DNS Servers UI Data © 2003 IBM Corporation Database Servers 2 File/Print Servers LAN Servers

IT Challenges Missing or Loss of critical data is immeasurable Up to 40% of today’s outages are unscheduled stoppages Complex, heterogeneous environments Poorly documented legacy applications make it painful to diagnose and resolve complex cross-product problems 25 -50% of IT resources are spent on problem determination and resolution The skills needed to do manual crossproduct problem determination are scarce and expensive Outages of mission-critical systems cost quite a bit Outages & unscheduled work leads to saturation on backup systems & power systems © 2003 IBM Corporation 3

Autonomic Vision “Intelligent” open systems that: § Manage complexity § Know themselves § Continuously tune themselves § Adapt to unpredictable conditions § Prevent and recover from failures § Provide a safe environment Providing customer value § Increased return on IT investment § Improved resiliency and quality of service § Accelerated time to value © 2003 IBM Corporation 4 Focus on business, not infrastructure

Autonomic Computing Attributes Self-managing systems that deliver: Increased Responsiveness Business Resiliency Discover, diagnose, and act to prevent disruptions Adapt to dynamically changing environments Operational Efficiency Secure Information and Resources Tune resources and balance workloads to maximize use of IT resources © 2003 IBM Corporation Anticipate, detect, identify, and protect against attacks 5

Levels of Autonomic Maturity IT Focus Business Focus Element Management • Manual processes • Reactive and tactically aligned • Resource intensive Centralized Management On Demand Management • Automated best practices • Proactive and Predictive • Autonomic, self managed environment • Automated provisioning • Dynamic provisioning Autonomic Adaptive Predictive 3 Managed Basic 1 © 2003 IBM Corporation 2 6 4 5 Evolutionary approach; Revolutionary outcome

Autonomic Computing Architecture The Big Picture of Autonomic Computing Technology • Define a base reference architecture model which creates a common vernacular for autonomic computing Business policy Resource Provisioning Workload Management Solution Install Autonomic core capabilities Problem Determination Admin Console Products delivering autonomic features • Deliver products with built-in autonomic capabilities • Create and leverage open standards for autonomic computing Open Standards © 2003 IBM Corporation Policy • Deliver core infrastructure technologies that provide for an open framework for the industry 7

Autonomic Computing Architecture Overview Sensors Analyze Monitor Effectors Plan Knowledge Execute Autonomic Manager Data Sensors Effectors Action Manageability Interface Resource Manager Managed Element © 2003 IBM Corporation 8

Autonomic control loops: next step evolution BUSINESS SLA POLICY USER RESPONS E TIME Autonomic

Policy and Autonomic Managers § Policy: a set of considerations designed to guide decisions on courses of actions (Calo, et al. ) Policy § In autonomic (self-managing) systems, policies reflect human judgement. Sensors § In an autonomic systems, policies are considerations, stored as knowledge, that guide the actions of autonomic managers Analyze Monitor § Policies can tell an autonomic manager what to achieve (goals) or how to achieve it (actions) Effectors Plan Knowledge Execute Autonomic Manager Policies allow administrators to configure autonomic systems. © 2003 IBM Corporation 10

Policy in Practice: e. WLM Workflow -centric management model èResource allocation based on goal achievement èService class goal exceptions trigger adjustment algorithms èPredictive algorithms avoid nonproductive adjustments èBusiness importance determines problem prioritization èBusiness importantance determines winners and losers Policy Based Business Priority based goals è 90% less than 2 seconds, critical to the business èAvg response time 3 seconds, important, but not critical èDefined by user group and/or business process Dynamic, learned topology controls èNo static definitions of servers/application environments Employee Scheduling Appliance Servers Web Application Servers Data and Transaction Servers Internet Business Partners Employee Benefits © 2003 IBM Corporation 11 Employee Scheduling

Policy in an IT Infrastructure Access control policies Quality of service policies Security policies

(Some) Key issues in policy for autonomic computing § Business driven: how autonomic manages can manage against high-level business policies. – Reduce the complexity by managing toward business objectives § Canonical specification: autonomic systems are heterogeneous. – Multiple policy specifications leads to overly complex, fragile systems. § Orchestration: how policies resolve conflicting systems needs. § Transformation, validation, user centered design, and many, many more… © 2003 IBM Corporation 13

Anatomy of an AC Policy: a canonical policy language § Four common concepts identified: – Scope – Specifications to identify what is or is not subject to the intent. – Precondition – Specifications to express when a policy is to be applied or is active. – Business Value – Specifications to express utility functions to make economic trade offs – Decision – Specifications to describe observable behavior or objective. § Designed by adopting concepts from prevalent policy languages – e. WLM, IETF/DMTF standard, WS-Policy, WSLA © 2003 IBM Corporation 14

AC Policies and WS-Policy defines a container for assertions. WS-Policy Document AC policies are domain assertions <wsp: policy> AC Policies Scope are assertions Precondition contained in WS-Policy documents Business Value Decision Scope Precondition Business Value Decision </wsp: policy> © 2003 IBM Corporation 15

Defining and Executing Policies Observations: § Policies are used by an autonomic manager to control a managed resource. § Therefore, policies must relate to the resource’s sensors and effectors. § At definition time, policies should be associated with a resource type § At run-time, autonomic managers apply the policies appropriate for the types of resource under management The next two charts illustrate these points. © 2003 IBM Corporation 16

Defining Policies S A 4 -tuple is a type of “K”. Scope: 4 -tuple Condition: E Autonomic Manager A M Business value: Decision: P K Policy 4 -tuple E List of Sensors Managed List of Effectors Resource’s descriptor Sensor Effector Managed Resource © 2003 IBM Corporation 17 The policy is a function of the MR’s externalized sensors and effectors. The managed resource’s descriptor enumerates a resource’s sensors and effectors

How autonomic managers work with policy Sensor Effector Autonomic Manager Analyze Monitor Plan CBE Knowledge Execute CBE Sensor Effector Managed Resource © 2003 IBM Corporation 18 Monitors for events. Deliver to “K””. “A” gets data from “K”. Retrieves state and consults K as needed for policy evaluation. Evaluates the policy conditions and business value Performs the command as indicated by the policy decision

an example – policy-driven self-optimizing solution: IBM Server Allocation for Web. Sphere High priority Stock Trading Load Balancer Web. Sphere Transactional Grid Mid priority Account Manager Low priority Forecaster Advice Application • Multiple Web. Sphere transactional Application Provisioning applications • Multiple Service Level objectives • Dynamic and automated application provisioning © 2003 IBM Corporation Parallel Services IBM Server Allocation for Web. Sphere Application Server v 5 19 Database Server

Mustang: Objective and Approach § Objective – Maintain service level objectives (e. g. response time) in the face of varying workload § Approach – Configuration management for rapid resource addition/removal – Online capacity planning to estimate resource needs – Policy enabled controller © 2003 IBM Corporation 20

Mustang: Policy enabled self-optimizing controller Sensor Analyze Optimizer Monitor Retrieve State Sensor WSLA Policy Server Agent Notification Execute Knowledge Command Managed Resource resp. Time, thruput Effector Start/Stop Server WAS 5. 1 SPEC driver WAS 5. 1 trade driver batch driver © 2003 IBM Corporation Plan Configuration Transition Engine Condition Evaluator Data Provider Effector Autonomic Manager WAS 5. 1 21 Server pool DB 2 8. 1

Sample policy on min nodes: determined by time-of-day During daytime (6 am 6 pm)

More sample Mustang time-of-day policies During nighttime (6 pm – 6 am) SPEC has

WAS Server deployment transition: night d day SPEC has 5 WAS nodes { SPEC

Unity system and policy issues § Unity is an autonomic system integration project § Focus issues: – – – Interactions between autonomic managers Conflict resolution Resource allocation Self-configuration (dynamic discovery and runtime binding) Self-healing § Policy focus: – Extending policy language and framework to support utility functions – Utility functions enable: – Principled policy decomposition – Conflict resolution – Continual optimization © 2003 IBM Corporation 25

Unity Architecture AC system composed of interacting autonomic components An Autonomic System Registry Sentinel Resource Arbiter Policy Repository Application Environment Demand Router Application Manager Router . . . Server Database © 2003 IBM Corporation 26 Storage Server Demand

Resource Allocation in Unity (utility-based approach) UI/Editor Resource Arbiter Policy Repository The utility functions provide The policy repository holds Scenario: an overall The application managers The arbiter allocates the ause principled basis for and distributes policies that solution contains a tell these policies to available resources in a way resolving the resource specify a business value number of applications, the resource arbiter the for that maximizes the expected contention, and the behaviors of for each and resources that must likely effects business value ofchanging the overall converting business-value application. be allocated among the allocations. solution. policies into action. applications. Demand Application Manager Server © 2003 IBM Corporation Application Manager Server . . . 27 Server

Unified Policy Management § Standardized approach to enable integration of policy- Policy Editing Tools based solutions § Independent of any product infrastructure § Rich syntax for the definition of policies § Persistence and distribution Policy of policy definitions Definition § Allow policy guidance to be (Policy Grammar) stated in a variety of forms (Results, Actions, and Goals) Decision Federator APIs S Preconditions: Log file size > 100, 000 entries A M Decision (Action): incremental backup © 2003 IBM Corporation 28 P K Decision Maker E Decision Maker APIs RM E Managed Element Business value: high E Autonomic Manager S Scope: database Decision Federator S RM E Managed Element

Research Challenges Policy: Set of guidelines or directives provided to autonomic element to influence its behavior. § Human Interface Ø Authoring and understanding policies Ø Avoiding or ameliorating specification errors § Developing a universal representation and grammar Ø Many different application domains, disciplines Ø Many different flavors of policy Ø Covers service agreements as well ? § Algorithms that operate upon policies (and agreements ? ) Ø Automated derivation of actions (e. g. planning, optimization) Ø Automated derivation of lower-level policies from high-level policies (e. g. maximize profit from this set of service contracts) § Conflict Resolution Ø Both design time and runtime Ø Need to establish protocols, interfaces, algorithms © 2003 IBM Corporation 29

The Journey Has Started § Products, technologies and services are available today § IBM is working with academia, business partners and standards organizations to develop open standards for self-managing systems § Broad IT industry participation is needed – this is an industry-wide initiative § Policy is a critical element and one of the most challenging aspects of AC § Innovation & Collaboration are a must!! Aggressive research is essential!! Freeing people to focus on their business instead of their infrastructure © 2003 IBM Corporation 30