Just-in-time Software Distribution in (A)IoT Environments

Abstract

Traditional software distribution systems are highly inefficient for the needs of Artificial Intelligence of Things (AIoT) devices. The processing power and other resources of modern AIoT devices enable the use of general-purpose operating systems (i.e., Linux) and thick stacks of libraries to implement specific functionalities at a high level of abstraction. However, these advantages do not come for free. General-purpose software is not inherently optimized in terms of performance and energy efficiency; a significant portion of resources is consumed for system functioning and maintenance; the complexity of the system potentially jeopardizes its stability and security, among other issues. However, one of the main drawbacks of this approach is the need for frequent software updates, which involves distributing a large amount of data to the devices and storing it on them. In this paper, we introduce a new approach to software distribution in the form of a just-in-time file-system model, which retains the functionalities of existing software management systems but significantly reduces the amount of data copied to the device (initially or during updates), thereby conserving resources and speeding up the update process. The research presented in this paper indicates that during software updates, up to 90% of files are unnecessarily replaced with identical copies. Therefore, by implementing the proposed file system, significant savings could be achieved in terms of communication channel usage, external memory capacity and durability, as well as processor time required for updates, although as a trade-off in system autonomy and dependence on network connectivity.