NextGeneration InternetScale Systems NinjaMillennium Projects

1
Next-Generation Internet-Scale SystemsNinja/Mill
ennium Projects

• David E. Culler
• Computer Science Division
• U.C. Berkeley
• ninja.cs.berkeley.edu
• www.millennium.berkeley.edu

2
Future Internet-Scale Systems

• Information Devices (10 Billion)
• Connected Stationary Computers (100 Million)
• Scalable Servers (Million)

3
Internet-Scale Systems

• Extremely large, complex, distributed,
  heterogeneous, with continuous and rapid
  introduction of new technologies
• Feasible architectures
• Decentralized, scalable algorithms
• Dynamically deployed agents where they are
  neededBig infrastructure, small clients
• Incremental processing/communications growth
• Careful violation of traditional layering
• Implementation approach based on incremental
  prototyping, deployment, evaluation,
  experimentation

4
Imagine

• You walk into a room

5
Vision ? Goal

• The next internet revolution will come from
  enabling component services and pervasive access
• not vertically integrated Information Technology
• ? Dynamic, programmatic creation / composition of
  scalable, highly available, customizable
  services
• Automatic adaptation to end device
  characteristics and network connectivity
• Arbitrarily powerful services on arbitrarily
  small clients using a proactive infrastructure

6
NOW Project

• Breakthrough design techniques for large-scale
  systems
• Low-overhead interface to enable emerging
  networks
• Active Messages
• Explore novel OS designs
• World-leading cluster
• Basis for Inktomi

7
Transcoding Proxies
Scalable Servers (NOW)

• Transformation, Aggregation, Caching, and
  Customization (TACC)
• Scalability and availability
• Limited customizability and locality and no
  persistence

8
ProActive Approach

• Create a framework that enables programmatic
  generation and composition of services out of
  strongly typed reusable components
• Key Elements
• Structured architecture with a careful
  partitioning of state
• Bases, Active Routers, and Units
• Wide-area paths formed out of strongly-typed
  components
• Operators and Connectors
• Execution environments with efficient, but
  powerful communication primitives
• Active Messages capsules
• TACC persistence customization

9
Structured Architecture
10
Example TopGun WingMan/Mediaboard Operation
AP
PDA
LegacyServer
PDA Proxy
Base
Image Converter Un-Zip
MediaBoard
PC
Multicast Connector
Aggregator
AP
Mic
Camera
11
NINJA Service Architecture

• Operators/Connectors/Interfaces
• Paths
• Wide-Area Paths
• Interface Interconnection
• Path Optimization
• Units, Active Proxies, and Bases
• Services
• Service Discovery
• Automatic Path Generation
• Example Applications
• Universal remote control/smart spaces
• Universal In-Box/Personal Information Management

12
Operators/Connectors

• Operators
• transformation
• aggregation
• agents
• PI provides secure execution environment
• Connectors
• abstract wires
• ADUs
• varying semantics
• uni/multicast

• Interfaces
• strongly typed
• language independent
• control channel
• path changes
• authentication
• feedback

13
Wide-Area Paths

• Creation (explicit or automatic)
• Query Service Discovery Service to find logical
  path of operators
• Place operators onto nodes
• Path is unit of resource allocation and
  authentication
• Connectors are polymorphic entire path must type
  check - statically
• Optimization
• Add (or transpose) operators
• forward error-correction
• compression/decompression
• Change operators, connectors, locations, or
  parameters
• Interoperability
• Wrapper operators for legacy servers
• Leverage COM objects as operators

14
iSpace Execution Environment
operator upload
Service request
service threads
Persistent Storage
Managed RMI
Physical processor
Operators
Caches

• parallel application framework on Bases
• RMI hides complexity of scalability and
  availability
• Dynamic customization and composition
• apSpace is limited execution environment for AR

15
Base Execution Environment

• Ninja RMI
• Sun RMI compatible serialization and thread
  management
• ninja remote object
• TCP or UDP or Multicast UDP (Active Msg soon)
• Authenticated public key
• iS-box
• customizable service VM
• Redirector
• iSpace

16
iS-box platform framework

• Loader Extends JVM to support services
• LoadService (URL, name, args)
• ListServices
• GetService(name) -gt svc obj
• KillService
• Trusted services loaded at startup
• Security MGR interposes on method calls
• loaded as a trusted service

iS-Loader
Trusted-Services
Security MGR
JVM
17
Push Services into the Infrastructure
New service
iS-Loader
Trusted-Services
Service Methods
RMI stubs
Security MGR
JVM
Generated by RMI compiler

• GetService returns service object
• Issue Programming Model for Service Methods

18
Scalable iSpace

• Multi-Space services across group of iS-boxes
• List, Get, or Load Service from any
• Get returns redirector stub

iS-Loader
Multi-Space Loader
Multi-Space SVC
Security MGR
JVM
19
Prototype Services

• iSpace Admin monitor
• automatic forms interface
• computational econ support
• juke box
• ninja fax
• pilot pager
• Digicash mint
• 1200 RMI/s on 400 PII w/ jit

20
Millennium Project Goals

• Enable major advances in Computational Science
  and Engineering
• Simulation, Modeling, and Information Processing
  becoming ubiquitous
• Explore novel design techniques for large,
  complex systems
• Fundamental Computer Science problems ahead are
  problems of scale
• Develop fundamentally better ways of
  assimilating and interacting with large volumes
  of information
• and with each other
• Explore emerging technologies
• networking, OS, devices

21
Campus-wide Millennium Testbed
22
Physical Connectivity
23
Resource Support

• Computers via Intel Technology 2000 grant
• 200 NT desktops
• 16 department 4-way SMPs
• 8 5x4 Group Clusters,
• 1 100x4 Campus Cluster
• PPro gt Pentium II gt Merced
• Additional storage via IBM SUR grant
• 0.5 TB this year gt 4 TB
• NT tools via Microsoft grant
• Solaris x86 tools via SMCC grant
• Campus provides Technical staff
• NorTel Discount
• NSF Research Infrastructure Grant

200 Gflop/s 150 GB memory 8 TB disk
24
High Speed Cluster Networking

• Transforms large collection of individual
  resources into a powerful system
• Emerging system area networks
• Myrinet, ServerNet, Synfinity
• Emerging user-level networking
• Virtual Interface Architecture
• Intel/Microsoft/Compaq std based on Active
  Messages and related research
• NT and Unix
• Gigabit Ethernet?

25
Inter-cluster NorTel network

• Gigabit Ethernet connecting group clusters and
  campus cluster
• Campus provides fiber plant, maintenance, and
  staff

26
Computational Economy

• How is this vast, integrated pool of resources
  managed?
• Traditional system approach empower global OS to
  provide optimal allocation to blind
  applications
• predefined metric, tuned to fixed workload
• ignores the inherent adaptation of demand
• Computer Center
• charge gt director-to-user feedback according to
  cost
• Economic view decentralized allocation according
  to perceived value
• pricing gt user-to-user feedback
• compatible niches,sense of control, cooperation
• Everything is a service!

27
Experiment PDA Bazaar

• Deploy/use pervasive computing infrastructure in
  Soda Hall
• Provide Ninja iSpaces
• Build an initial community (200 PP3 / Workpads)
• Watch and evaluate
• information broadcast channels
• seminars, lecture content
• news/sports/stocks
• shared information
• calendars, room reservations
• collaborative note-taking and brainstornming
• Smart spaces and device control

28
Perspective

• Computer Science Research Challenges Ahead are
  fundamentally problems of scale and composition
• Powerful services
• Billions of tiny devices
• Rapid development, deployment, and evolution
• These demand an entirely new approach to systems
  research.
• Academic / Industry collaboration is key