The Workshop on Internet-scale Software Technologies

Organizational and Technical Issues in the Tension Between
Centralized and Decentralized Applications on the Internet

July 13-14, 2000
University of California, Irvine
Irvine, California, USA

Welcome to TWIST 2000!




Important Dates

Lodging Information

Directions and Parking


Organizing Committee

For More Information

TWIST Series


Sponsored by:

UC Institute for
Software Research (ISR)


» The attendance list is now available.

The goal of TWIST 2000 is to substantively explore design tensions between centralizing and decentralizing forces on the Internet, the pros and cons of centralized and decentralized architectures, and the long term implications which lead architects to design one way or the other.

Many of the most successful applications on the Internet today are architecturally centralized. Among these are eBay, AOL, and The success of these centralized architectures is surprising to some, given the fundamentally decentralized way the Internet itself and the World Wide Web work.

Alternatively, many companies and research projects have advocated decentralized applications. Such applications are touted as having the advantages of robustness, scalability based upon replication (rather than just raw speed), resource sharing, and ability to span trust domains. Applications of the decentralized approach include SETI@Home (parallel scientific computing) and the Air Traffic Control system (distributed command and control).

Many applications employ a mixed strategy, including financial trading and email. Consider how Travelocity, for example, is implemented as a decentralized Web application wrapping the centralized Sabre reservations service. Other applications exhibit both strategies depending on the layer of abstraction considered: the Domain Name Service is a centralized monopoly of names in a decentralized database, or how Akamai appears as a single global Web cache to a browser but internally relies on globally distributed servers, or eBay, a centralized service enabling wildly decentralized marketplaces.

We seek answers to such questions as:

  • Can centralized applications continue to scale with the growth of Internet users, traffic, types of services, and customer base?
  • Can existing centralized approaches continue to grow unabated, or will they reach hard limits?
  • If they can grow unabated, then how can this be accomplished and how does it impact decentralized application architecture and development?

Issues to consider include:

  • At what levels of an application's design should distribution be employed?
  • What are the key distinguishing characteristics of services (applications) for which centralized architectures (exploiting Moore's Law) will continue to suffice?
  • Under what circumstances are decentralized architectures superior? Necessary?
  • What sort of application spaces do applications such as Internet phones/smartphones have?

Axes of influence include:

  • Economic and business models.
  • Trust.
  • Robustness/fault-tolerance.
  • Scale.
  • Problem characteristics.
  • Democratization. Participants often vote their resources by deciding to share information or compute cycles.



Attendance at the workshop was by invitation only, based on submission of an informal statement of interests.

A list of TWIST 2000 attendees is available.

Workshop Report

The workshop organizers will produce a report subsequent to the workshop which wil be submitted for widespread publication. A proceedings will not be produced.

TWIST Workshop Series

TWIST 2000 is the third workshop in the annual TWIST series. TWIST 99 focused on Internet-scale Namespaces. WISEN 98 focused on Internet-scale Event Notification.


Sponsored by the Institute for Software Research
Send comments to