In OpenTelemetry metrics, the choice between Delta and Cumulative temporality involves significant trade-offs related to data reporting and backend compatibility. Delta temporality records changes between reporting intervals, while Cumulative temporality aggregates data from the start of the measurement. The decision on which temporality to use often hinges on the capabilities of the backend systems, with some, like Prometheus, favoring Cumulative due to its resilience against data loss from dropped samples. The OpenTelemetry Collector can convert between temporalities, but scaling such conversions poses challenges, particularly when addressing issues like out-of-order writes and data aggregation. Cumulative temporality typically requires more memory for Synchronous Instruments due to storing every metric-label combination, whereas Delta temporality can minimize memory usage and reduce the risk of memory leaks. The low-memory mode of OpenTelemetry SDKs balances this by using Delta for Synchronous and Cumulative for Asynchronous Instruments. Ultimately, the best practice is to align temporality with backend support, with Cumulative often recommended for its robustness against data gaps and simplified pipeline management.