CAREER: Towards Efficient In-storage Indexing

Data indexing plays a crucial role in numerous modern technologies, including search engines, big data analytics, file systems, and databases. In this context, in-storage indexing devices (ISIDs) have emerged to enhance the functionalities of storage devices, leading to improved performance, efficiency, and cost-effective data processing. By storing index information alongside the data it indexes within the same storage device, ISIDs offer several advantages over traditional indexing methods.

These advantages include reducing data movement, improving access speed, minimizing network impact, enabling efficient data management, and freeing host computing for critical tasks. To design efficient ISIDs, several challenges need to be addressed. Firstly, there is a need for low-cost and open-source research platforms to facilitate the reproduction and comparison of research work, promoting quick adoption of ISID advancements.

Secondly, integrating the fragmented advancements of individual ISID components is crucial to capture their holistic impacts and interactions effectively. Thirdly, addressing diverse workload requests, interference in multi-tenant environments, and data distribution considerations requires new research methods for overall operation optimization. This CAREER research project aims to overcome these research challenges and promote the adoption of ISIDs, contributing to the advancements of storage systems.

This project will explore and develop innovative methods to unleash the full potential of ISIDs in modern data management systems. By addressing the core challenges, the project seeks to revolutionize data storage systems and make significant contributions to the field of storage technology. This project will share the findings with undergraduate and graduate students through computer science programs and open up career opportunities to female students, underrepresented minorities, and first-generation college students.

This project will disseminate the proposed techniques into the industry and foster technology transfer through new industrial collaborations. The developed infrastructure will be available to the research community through a web-based portal.

This research makes significant empirical contributions to the ISID design and development space by addressing major challenges posed by in-storage indexing. Specifically, it advances the state of knowledge by investigating the following questions: (1) How can we design and develop new ISID models that accurately capture the behavior of internal modules, such as the index manager, request handler, data access parallelism, index-induced wear leveling, and garbage collection?

These insights will enable scientific design advancements and detailed tradeoff analysis for ISIDs. (2) How can we develop efficient dynamic model calibration techniques using coarse measurements to parameterize queuing models that accurately capture burstiness and variability in ISIDs? (3) How can we emulate index manager delays using different data structures and sizes and utilize black-box and gray-box calibration techniques to identify ground truth for ISIDs? (4) How can we design a new re-configurable indexing architecture and index cache that ensures deterministic tail latency, low overhead prefetching and eviction, and improved membership checking through object signatures and ML-based feature learning in ISIDs? (5) How can we design tenant-local eviction policies that consider the effect of allocating space for index and data, recognizing the dependencies between them for efficient data access in ISIDs? (6) How can we minimize log-checking overhead and avoid in-storage hash computations while exploring the trade-off between consistency and performance by allowing read-only tenants to bypass the log and access their own consistent index in ISIDs? (7) Does capacity variance, which gracefully reduces ISID capacity as flash pages become bad, provide a better alternative to wear-leveling for ISIDs? Throughout the project, the PI will facilitate the connection of the proposed research with the contents and concepts of several courses on Systems at FIU.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.