Much of Ethernet's strength lies in its simplicity. Computers on an Ethernet communicate via a single passive broadcast medium, with no central controller. Each computer simply requires an Ethernet interface and a connection to the network. The flow of data is cooperatively controlled by all of the computers on the system, and the communication medium is completely passive. Autonet, however, was designed with performance as its primary goal, requiring concurrent packet transmission and multiple packet pathways in order to provide increased bandwidth. Thus, the Autonet network is an active, controlling agent composed of a number of complex switches inter-connected via point-to-point links. Each individual switch plays an active part in the passage of data through the network. Thus, the responsibility for control of the data has been moved from the connected computer systems onto the network itself, greatly increasing its complexity. Just the crossbars within the switches are far more complicated that Ethernet's single wire, and add significant control problems. For instance, in Autonet flow control between and among switches must be addressed, and requires the capability to transmit and process control signals (as distinct from packet data) between switches. Flow control problems such as avoiding deadlocks also introduce added software complexities. These issues require such complexity that a dedicated microprocessor is required in each switch.
All of the extra added hardware in Autonet also complicates problems associated with hardware failures. In Ethernet, the interface of each computer is designed to remove itself electrically from the network in the event of a failure. Thus, a single fault does not bring the entire network down, provided that the interfaces disable themselves correctly. Autonet, however, is not nearly so simple. A failure in Autonet requires more than disabling the faulty component, since the topology of the network has been altered. Autonet must re-configure itself, re-setting the routing data stored in all of its switches. This is a complex process, and requires a shutdown of the network for a brief period of time. It also requires the aforementioned capability of the switches to communicate control data among themselves, such as the ability to check on the status of its neighbors. Furthermore, the numerous switches and links of Autonet are far more likely to produce a failure than the simple wire backbone of Ethernet, so that reliability pays a price for the increased complexity.
Despite Autonet's added complexity, it possesses a number of important advantages over Ethernet. The bandwidth of Autonet is much greater than Ethernet, and can be easily increased by adding switches and links to the network. Automatic re-configuration allows arbitrary network topologies which are not possible on Ethernet. Autonet also uses `pushback' flow control and buffering to minimize packet losses, whereas Ethernet simply discards packets. Autonet thus proves to be a far more powerful network design than Ethernet. However, at the present time, Autonet's advantages probably only outweigh its cost in applications that are performance-critical. In a system where the extra bandwidth of Autonet is unlikely to be used, Ethernet is likely the better choice due to its reliability and low cost. Nevertheless, in the near future Autonet's high performance will likely outweigh its costs as the fast-paced increase in computing power outstrips the usefulness of Ethernet.