The Design of a Dynamic Network for Enterprise Systems

1
The Design of a Dynamic Network for Enterprise
Systems

• Bina Ramamurthy
• Major Area Distributed Systems

2
Introduction

• After 20 years in academics and the Silicon
  Valley, the new Provost of Stanford University
  calls for a shift in focus for systems research.
  Performance -- long the centerpiece -- needs to
  share the spotlight with availability,
  maintainability, and other qualities.
• Foreword for John Hennessys article Future of
  Systems Research, CACM, August 1999.

3
The Big Picture

• We are fortunate to be participants in an ongoing
  technological revolution. Three catalysts for
  this revolution are desktop, network and
  enterprise. Tremendous advances have taken place
  in these areas.
• A major goal of my research is to design a novel
  architecture for enterprise-wide distributed
  computing that will leverage the advances in
  these areas to satisfy the ever increasing
  demands of todays application domains.

4
Topics for Discussion

• Enterprise Systems
• Issues in Enterprise Systems
• Traditional Solutions
• Our Approach Dynamic Network
• Enabling Technologies
• Research Problems

5
Enterprise Systems

• An enterprise is a very large organization.
• An enterprise system is a distributed system
  involving many large organizations.
• An example ATT, inktomi, amazon.com, UPS, and
  users operating in a supply chain model, make up
  an enterprise system.
• Inter .com .

6
Enterprise Computing
E1
E2
E3
Enterprise Level
P1
P2
D1
D2
D3
Project/ Department Level
W11
W23
W22
W21
W12
Workgroup Level
I32
I11
I21
I22
I31
Adapted from Distributed Information Systems
in Business by Konig et al., Springer Pub.
7
Evolution of Computing Systems
BR
Enterprise Systems
Distributed Systems
Client

Centralized Systems
/Server Systems
8
Distributed System as an Enterprise System

• There are many problems in using traditional
  distributed system model for enterprise
  computing. Look at
• A Note on Distributing Computing by Jim Waldo,
  Geoff Wyant, Ann Wollarth and Sam Kendall of Sun
  labs.
• -- current distributed system paradigm works well
  for small systems dealing with single address
  space but fails very badly for dynamically
  changing global address spaces.
• We have seen advances in code mobility, data
  mobility,etc. But the distributed system
  architecture/principles are yet to evolve in any
  significant way.
• Focus on distribution.

9
Issues in Enterprise Systems
BR
APPLICATION
Ease of use Uniform interface Design and
development effort Flexibility Rapid Application
Development (RAD)
Definition of a Model Distribution Scalability Ava
ilability Load Balancing Security Interoperability
Server Power
Response time end-to-end QoS User Interface
SYSTEM
Return of Investment Total Cost of
Ownership Design to Production
Time
USER
BUSINESS
10
Requirements for Enterprise Computing

• Accommodate changes gracefully - scalability,
  dynamic reconfiguration
• Maintain high availability at all times
• Offer good performance in terms of response time
  and end-to-end QOS
• Dependability and fault tolerance
• Simplicity
• .

11
Traditional Solutions

• Redundancy just as it is in a space mission
  backups.
• Redundancy with some policy Triple modular
  redundancy.
• Problems
• All these are age old solutions with no focus
  inter-system/space computation.
• Redundancy may not be possible in an enterprise
  where clients are growing thinner and thinner.
• Redundancy alone may not solve the problem in an
  entity which can be a server as well as a client.

12
Our Premise

• Raw redundancy alone will not help. We need
  effective redundancy management.
• A highly efficient distribution mechanism is
  imperative for high scalability and availability.
• Most currently available control services (naming
  etc.) in a distributed environment are necessary
  but are quite rigid and static.
• Application service providers can be any device,
  hardware /software, small or large.

13
Our Approach

• Guiding principle
• - Convergence of desktop, network and enterprise.
• Client level desktop refers to all services
  on the network
• Infrastructure level network
• Application level enterprise
• A dynamic network is defined that focuses on
  dynamic distribution of services. Architecture
  and design details are explained next.

14
Architecture

• The functional requirements of the system are
  classified into two categories
• Static and dynamic
• Dynamic network of the system is strongly rooted
  in a static structure.
• Static network handles functions such as
  naming, lookup, discovery, remote events,
  transactions.

15
Dynamic Network

• Dynamic network refers to all those components
  whose configuration and contents change
  dynamically.
• This is the main contribution of this research.
• This is the component which distinguishes our
  work from the existing distributed system
  technology.
• It supports two new concepts distribution
  service and publishing service.

16
Distribution Service

• Simple redundancy at all service providers is not
  going to solve the problem of availability,
  scalability or effective distribution.
• Primary goal of the distribution service is to
  provide an optimal distribution of services among
  the participants. It relies on the publishing
  service and the static network to gather
  knowledge about the available services.

17
Publishing Service

• Each participant uses this service to publish the
  availability of the services.
• Many service related attributes such as qos,
  response time, etc. are included in the
  publication.
• Distribution service uses this information to
  dynamically configure the network for optimal
  availability.
• Addition of services and servers and degradation
  or crash of a server can be elegantly handled by
  distribution service.

18
Deliverables
Dynamic Network Definition (publishing service,
distribution service)
Time to delivery is critical.
Static Network Definition (protocols, technologies
, operational details)
19
Enabling Technology
BR
server
server
client
client
desktop
middleware
middleware
network
20
Motivation

• The Next Generation Internet (NGI) must have a
  whole new set of characteristics and powerful
  features. Most importantly, it should be based on
  standard, completely open middleware.
• Dr.Schmidt, an eminent researcher in this area,
  in his research statement.
  
  
  
  
  
  

21
Middleware Technology

• OMGs CORBA standard language and platform
  independent.
• SunMicro Systems Jini for Java-enabled world.
• Microsofts DCOM Proprietary.
• Others

22
Research Problems

• A Model (Design) for Highly Scalable and
  Available Enterprise Systems (Ph.D.)
• Design, Analysis and Implementation of
  Distribution Service (Ph.D.)
• Design, Analysis and Implementation of Publishing
  Service (M.S.)
• On the application design end Component Oriented
  Design (COD) methodology (M.S.)
• An Adaptive Robotic Community Framework
  (JiniBots)
• Jointly with Matt Stock of CIT, SUNY Buffalo