In this work, we present a set of comprehensive experimental studies on Hyperledger Besu in private blockchain. We aim to exhibit its performance characteristics in terms of transaction throughput, latency, resource utilization, and scalability, from the application perspective by adding a load balancer middleware. We have carefully designed a set of comparative experiments and judiciously selected typical parameters, including transaction send rate, network size, node flavor, load balancing, consensus, and block time. In particular, three proof of authority consensus algorithms, Clique, IBFT 2.0, and QBFT, are investigated. Through extensive experimental evaluations using the Hyperledger Caliper benchmark tool, we analyze how these parameters impact the performance of a private Besu blockchain. Our studies reveal several interesting findings. 1) Blockchain parameters, e.g., block time and block size, are the most significant factors in determining Besu performance; 2) The performance of Besu is bottlenecked by transaction execution and blockchain state updates, which are determined by parameters such as node computation power, transaction complexity, and load balancing; 3) A Besu network with QBFT consensus can scale up to 14 validators without noticeable performance loss. Our findings shed some light on further performance improvement of Hyperledger Besu. The identified bottlenecks and root cause analysis provide insightful suggestions for blockchain practitioners to build performant enterprise applications.