A Cross-Layer Decision Framework for High-Performance Cross-Platform Front-End Applications

Abstract

Front-end performance often involves cross-cutting decisions where trade-offs must be made across layers of architecture. However, the justification for a trade-off is often made in isolation. The cost of one optimization is pushed to the server, the network, the device, or the monitoring pipeline and is often not measured. The paper identifies when and where performance cost moves between rendering, server-assisted delivery, state, caching and prefetching, observability, and platforms. The paper also asks how to observe cost movement in practice and what measurable signals indicate it. A structured search of peer-reviewed journal and conference literature was conducted, supplemented by backward citation checking for foundational work. Each source was verified against the publisher's bibliographic record and assigned to one of four evidence classes. A cost-transfer tracing exercise, using a hypothetical media-rich interactive commerce application as a recurring illustration, synthesizes five distinct repeated patterns related to pushing costs around layers: rendering pushes costs between the client tier and the server tier, aggregation pushes round trips into the server tier, and prefetching pushes costs into the speculative future. Caching reduces repeated latency but transfers risk to freshness and invalidation; telemetry improves visibility but consumes runtime and network resources. The performance cost of state management matters more concerning the update scope and subscription granularity than the library choice. Cross-platform overhead is dependent on metric, engine, and execution model, not framework name. The strongest evidence (learned preloading short video feeds) shows wasted bandwidth is a first-class metric alongside quality of experience. From the evidence gathered, a decision framework is developed that relates performance goals to measurable signals, architectural levers, and the cost each lever would most likely transfer. Consequently, it is believed that front-end performance is a conditional, cross-layer property of the whole system, and an improvement can be trusted only when the layer that pays the cost is likewise measured.