Paper Title: Challenges in Designing Key Value SSDs

Paper Abstract: With the rapid rise in big data being generated by tech companies such as Google, Facebook, and LinkedIn, more data is being organized as Objects and stored in the cloud as Key-Value Stores. Key-Value stores are prevalent in web indexing, caching, storage systems, and Blockchain, and are in action in RocksDB, MongoDB, and LevelDB Software stacks. Traditional SSDs use a block driver interface to communicate with the host that requires applications data to first get converted to block interface from the object using host OS, Driver, and File systems, impacting system latency. Key-Value SSD designs have Key-Value stores inside of SSDs, and thus avoid the object-to-block interface conversion, but this is not yet successful due to various SSD design challenges. Relational Databases were all built on top of B+ Trees. B+ Trees have O(nlogn) search and write time complexity that doesn’t help Key-Value DB use cases that demand low write latency. Key-Value Stores uses two LSH and LSM trees to optimize for write-heavy updates done in O(1) and O(log(n) time. The challenges in designing a key-value SSD using any data structure design are Limited DRAM, CPU resources on the SSD, and variable size Key-value pairs.

Paper Author: Vishwas Saxena, Technologist, Firmware Engineering, Western Digital

Author Bio: I am Senior Technologist, Firmware Engineering at Western Digital, where I am both project lead and architect for NVMe-based CFexpress removable cards, wireless storage Drives, Edge Analytics based Video Surveillance Camera system. I have been a leader in creating new product lines and designing and debugging complex software systems such as flash firmware. I have more than a decade of experience in firmware development and more than two decades of experience (22 years) in the technology industry. I have a strong background in embedded software, device drivers, storage, security, networking, and wireless systems. I earned a Master's in Science from LJMU, UK in Machine Learning and AI (2021), and a Bachelor’s in computer science (2000) from the Netaji Subhas Institute of Technology (Delhi, India)