Homomorphic encryption allows arithmetic operations to be performed on ciphertext and gives the same result as if the same arithmetic operation is done on the plaintext. Homomorphic encryption has been touted as one of the promising methods to be employed in Smart Grid (SG) to provide data privacy which is one of the main security concerns in SG. In addition to data privacy, real-time data flow is crucial in SG to provide on-time detection and recovery of possible failures. In this paper, we investigate the overhead of using homomorphic encryption in SG in terms of bandwidth and end-to-end data delay when providing data privacy. Specifically, we compare the latency and data size of end-to-end (ETE) and hop-by-hop (HBH) homomorphic encryption within a network of Smart Meters (SMs). In HBH encryption, at each intermediate node, the received encrypted data from downstream nodes are decrypted first before the aggregation, and then the result is encrypted again for transmission to upstream nodes. On the other hand, the intermediate node in ETE encryption only performs aggregation on ciphertexts for transmission to upstream nodes. We implemented secure data aggregation using Paillier cryptosystem and tested it under various conditions. The experiment results have shown that even though HBH homomorphic encryption has additional computational overhead at intermediate nodes, surprisingly it provides comparable latency and fixed data size passing through the network compared to ETE homomorphic encryption.