Optimizing Edge Telemetry Synchronization: Benchmarking Durability Primitives and Micro-Batching Resilience in Resource-Constrained Environments

Edge deployments in remote areas frequently suffer from network partitions, dropped packets, and limited flash storage write cycles. While SQLite offers simple embedded ACID storage, the overhead of strict write durability during offline periods and subsequent recovery across degraded links is rarely measured together. In this work, we benchmark SQLite Write-Ahead Logging (WAL) settings and micro-batch network recovery over both clean and degraded links (250 ms delay, 15% packet drop). We observed that per-sample disk syncs (PRAGMA synchronous = FULL) impose a 359x write penalty compared to NORMAL mode. For egress, single-record streaming (N = 1) under degraded conditions caused synchronization time to spike from 1,038.9 ms to 228,232.6 ms (~3.8 minutes), triggering 15 socket timeouts. Grouping records into micro-batches of N >= 50 cut synchronization time down to 2,069.0 ms (a 99.1% drop) and eliminated dropped connections entirely.