Boosting Key-Value Stores: HeterMM Leverages DRAM for Enhanced Performance

Emerging technologies like NVM offer a cost-effective alternative to DRAM for key-value stores, but their performance limitations remain a challenge. HeterMM, a new framework developed by Xuan Zhou’s research team, aims to bridge this gap by utilizing DRAM’s superior performance to optimize data access and storage. HeterMM strategically stores frequently accessed data in DRAM, while utilizing NVM for long-term storage. This approach minimizes access latency and improves overall performance. The framework also employs a data storage mechanism and an operation log for data persistence and recovery, further enhancing its reliability. Extensive tests demonstrate that HeterMM outperforms existing solutions, making it a promising option for building faster and more efficient key-value stores.