Simplifying Management Hank Levy Doug Terry Vivek Pai

Simplifying Management Hank Levy, Doug Terry, Vivek Pai, Peter Druschel, Doug Comer, Karsten Schwan, Mike Reiter, Remzi Arpaci-Dusseau

Basic Problems • • • We’ve created a monster People hate computers (for good reasons) Stuff doesn’t work right It’s not intuitive Components don’t interact properly Confusing when things don’t work (what’s wrong? how do we fix it? ) • Difficult to manage in all environments Computers are a major cause of stress for a huge number of people in our society, both at work and at home

Environments • • • Mobile / embedded The home (unmanaged) The enterprise (IT managers) The network …

The Home Environment • In the home we have: – Heterogeneous hardware devices • computers • a/v and entertainment devices • control devices (lights, heating, security) – Heterogeneous software systems • commodity OSs • “hidden OSs” (in my Xbox or TIVO) • embedded control software (thermostat, …. ) – Variety of services provided by remote entities • backup • entertainment • monitoring

Poor Aunt Betty • • How does she set firewall policy? How does she control object access (files, devices)? How does she connect her TIVO to her PC and stereo? How does she manage the software life cycle? – installation, configuration, update, retirement, migration • How does she know what content is coming into and out of the house (wifi, cable, dsl, phone, cell phone…)? • How does she get a high-level view of what’s in her house • How does she share with friends and family? • How does she limit the boundaries of her house (where does her house end and her neighbor’s begin)? • What if she moves to a new house?

The Enterprise Environment • • • Large number of computers Large number of software systems Large number of devices and sensors Different versions / generations of SW and HW systems Complex network infrastructure (routers, firewalls, etc. ) Complex structures built from interacting off-the-shelf HW and SW components Distributed integration across geographic locations Complex sharing relationships with other enterprises Mobile users and devices outside of the firewall Business and legal constraints

Poor Fred (the IT manager) • Difficult to set policies • Hard to ensure that policies reflect business rules • Hard to ensure that policies are consistently applied across organization • Hard to manage the software life cycle – Installation, configuration, update, retirement, migration, on a collection of heterogeneous computers • Hard to specify and manage access control • Managing real-time data acquisition and analysis • Problem diagnosis across heterogeneous (possibly badly interacting) software components • Impossible to test infrastructure at enterprise scale • Hard to ensure that we’re meeting contractual agreements, SLAs, etc?

Vision • Aunt Betty should be able to buy a bunch of devices at Radio Shack, bring them home, plug them in, and have them work seamlessly along with all of the other devices already in her house, without any explicit configuration. • Fred should be able to easily create policies and verify that they meet requirements. Software life cycle should be easily manageable across the enterprise. Should be possible to ask “what if” questions without actually installing or changing hw or sw.

Some Possible Research Tasks (1) • Tools for easily creating and verifying policies • Automated tools for driving configurations from policies • Tools for object access control in complex heterogeneous environments • Models and tools for verifying assumptions about the environment • Automated component installation (no wizards/installers) and life cycle mgmt • Global visibility, control, and coordination of components (how do I reset the clock in each of the 50 devices in my house? ) • Better protocols for devices to learn about each other and interact

Some Possible Research Tasks (2) • Tools for collaboration across houses or enterprises – creation of virtual communities or enterprises • “What if” tools – tell me whether something I’m going to do will work – if I make this policy change, what will it cost in infrastructure? • “Fix it” tools – e. g. , if something is broken, tell me what I need to do – or better yet, just fix it and then tell me what you did • Automated instrumentation and monitoring • Fault and event analysis plus correlation of events • Automatic migration of systems, data, configurations, …. • Time travel across multiple hw/sw systems (please make it work like it did yesterday!)

Infrastructure Needs • Open edge devices for experimentation • Multiple “home-like” environments with large numbers of computing, a/v, and control/sensor devices • Data centers and routers to experiment with services • Enterprise-scale systems for testing of tools • Workloads to drive testing • Huge data pipes between enterprises

In the end… • We need to learn how to construct components that by design: – allow simplified and flexible interaction with other components – are robust and reliable – are easily monitored – are easily controlled – are easy to update