Abstract Serving DRAM as the storage through key–value abstraction has proved as an attractive option, which provides fast data access for data-intensive computing. However, due to the drawbacks of network round trips and requesting conflicts, remote data access over traditional commodity networking technology might incur high latency for the key–value data store. The major performance bottleneck lies in client-side request waiting and server-side I/O overhead. Accordingly, this paper proposes RHKV: a novel R DMA and H TM friendly k ey– v alue store to provide fast and scalable data management by using the designed G-Cuckoo hashing scheme. Our work expands the idea as follows: (i) An RHKV client transmits data requests to our improved Cuckoo hashing scheme — G-Cuckoo, which constructs a Cuckoo graph as directed pseudoforests in RHKV server. The RHKV server computes the reply for each data request. The server maintains a bucket-to-vertex mapping and pre-determines the possibility of a loop prior to data insertion. Through the use of this Cuckoo graph, the endless kick-out loop of data insertions that can potentially be experienced in the case of generic Cuckoo hashing can be detected. (ii) Despite messaging primitives are slower than RDMA READs for data requests, RHKV adopts RDMA messaging verbs unconventionally. It leverages rich network semantics and makes full use of RDMA’s high bandwidth and low latency for data access over high-performance RDMA interconnects. (iii) Moreover, in order to ensure the data operation’s atomicity, RHKV strives to utilize the advanced HTM technique. Experimental performance evaluation with YCSB workloads shows that, when basic data operations are conducted, RHKV outperforms several state-of-the-art key–value stores with lower latency and higher throughput.