Introduction

In the 1960s and 70s when the core ideas underlying the Internet were developed, telephony was the only example of successful, effective, global-scale communications. Thus while the communication solution offered by TCP/IP was unique and groundbreaking, the problem it solved was telephony’s: a point-to-point conversation between two entities.

As we move into an era increasingly dominated by Artificial Intelligence, new features and requirements grow out of it. We estimate that AI applications will account for more than 80% of all computing power used around the world. This increased demand significantly strains the existing network stack and brings heavy financial pressure to AI service providers and users.

While today’s internet has exceeded all expectations for facilitating ubiquitous interconnectivity, it is overwhelmingly used for computationally intensive applications. Just as the telephone system would be a poor vehicle for the broadcast content distribution done by TV and radio, the Internet architecture is a poor match to its primary use today.

Thus, the computing power one would want for AI and computationally intensive applications is made unnecessarily hard by the current host-to-host model of the internet, often requiring awkward or expensive workarounds. Given this discordance between historical design (host-oriented communication) and current usage (ubiquitous computing access), we ask: what would the architecture look like if we built it around the access of computing power?

As our project title suggests, we propose to generalize the internet architecture by networking the computing power. This change to the hourglass model, allowing the thin waist to use computing power as common utilities, makes computing power rather than its containers a first-class citizen in the Internet architecture.

Figure 1. Thin waist of computing network

CONA is a new network architecture that is based on network, focused on computing, deeply converging AI, Blockchain, Data and Security. It improves computing power retrieval and intelligence service access by providing stronger and more architecturally coherent support for persistence, availability, and authentication. The goal is to provide social-level services featuring global access and instant availability to put together the implementation of ubiquitous computing power, network infrastructure and artificial intelligence.

We emphasize that the CONA model is compatible with today’s Internet and has a clear, simple evolutionary strategy. Like IP, CONA is a “universal overlay”: CONA can run over anything, including IP, and anything can run over CONA, including IP. IP infrastructure services that have taken decades to evolve, such as DNS (potentially ENS) naming conventions and namespace administration or inter-domain routing policies and conventions, can be readily used by CONA.

Indeed, because CONA’s hierarchically structured computing resources are semantically compatible with IP’s hierarchically structured addresses, the core IP routing protocols, BGP, IS-IS and OSPF, can be used as-is to deploy CONA in parallel with and over IP. Thus CONA’s advantages in computing power distribution, application-friendly communication and naming, robust security, and protected privacy can be realized incrementally and relatively painlessly.

The next sections present CONA’s architectural design and implementation details.