[Paper Review] Locally Served Network Computers

Jim Gray|ArXiv.org|Sep 2, 1998

Quantum Computing Algorithms and Architecture3 citations

TL;DR

This paper proposes a decentralized model for Network Computers (NCs) where each office or home hosts a local server to handle high-bandwidth NC workloads, avoiding reliance on expensive, high-bandwidth WAN links or centralized servers. By deploying low-cost, locally managed servers, the approach reduces infrastructure costs and improves responsiveness compared to centralized models, offering a practical evolution of the PC paradigm for ubiquitous computing.

ABSTRACT

NCs are the natural evolution of PCs, ubiquitous computers everywhere. The current vision of NCs requires two improbable developments: (1) inexpensive high-bandwidth WAN links to the Internet, and (2) inexpensive centralized servers. The large NC bandwidth requirements will force each home or office to have a local server LAN attached to the NCs. These servers will be much less expensive to purchase and manage than a centralized solution. Centralized staff are expensive and unresponsive.

Motivation & Objective

• Address the impracticality of relying on high-bandwidth WAN links and centralized servers for Network Computers (NCs).
• Overcome the limitations of centralized server models, which are costly and unresponsive.
• Propose a scalable, cost-effective alternative for deploying NCs in homes and offices.
• Enable ubiquitous computing through localized, low-cost server infrastructure.
• Provide a practical evolution of the PC model for widespread, efficient deployment of networked computing devices.

Proposed method

• Deploy a local area network (LAN) with a dedicated, low-cost server in each office or home environment.
• Use the local server to host and manage NC workloads, reducing dependency on remote centralized servers.
• Distribute computing resources closer to end-users to minimize latency and bandwidth usage.
• Leverage existing LAN technology to provide reliable, high-throughput connectivity for NCs.
• Design the system to be self-contained and manageable at the local level, reducing administrative overhead.
• Position the local server as the primary processing and storage node for NCs in a given environment.

Experimental results

Research questions

• RQ1 Can a decentralized server model reduce the reliance on high-bandwidth WAN links for NC deployments?
• RQ2 Is a locally hosted server architecture more cost-effective and responsive than centralized server models for NCs?
• RQ3 What are the performance and scalability implications of deploying a local server per office or home for NC workloads?
• RQ4 How does local server deployment compare to centralized models in terms of manageability and operational cost?
• RQ5 Can local servers enable a practical, scalable evolution of the PC model toward ubiquitous computing?

Key findings

• Local servers significantly reduce the need for high-bandwidth, expensive WAN connections to the Internet.
• Local server deployment is more cost-effective and easier to manage than centralized server solutions.
• The model improves responsiveness by reducing latency through proximity of computation and data.
• The approach enables a scalable, practical deployment model for Network Computers in real-world environments.
• Local servers provide better fault isolation and operational autonomy compared to centralized architectures.
• The model supports the vision of ubiquitous computing by making NCs more accessible and efficient in everyday settings.