Smart-grid applications have widely varying data needs as well as bandwidth and latency requirements. The usual approach to collecting the available data (e.g., from Phasor Measurement Units) at a centralized site continuously and executing all the applications there leads to large latencies and requirements for high communication bandwidth. This paper proposes and evaluates, using real data, techniques wherein the data packets are disseminated based on the applications' data needs and semantics. These techniques systematically filter data in the dissemination network and reduce bandwidth requirements while resulting in low latency solutions. For example, based on the results from our testbed for the Indian Electric Grid, we show that the processing overheads decrease by at least 50% for the large PMU data sizes compared to the traditional centralized approach.