NSF-BSF:CIF:Small:Reliable Data Storage on Sampling Channels

University Of California-Los Angeles, Los Angeles CA

Investigators

Abstract

Explosive data growth and insatiable demand for data processing and computing have created an urgent need to develop sophisticated information systems able to timely handle advanced data processing requests. Emerging solutions increasingly employ more complex, hyper-scaled, decentralized data storage systems and sub-systems, ranging from the multi-cloud to block-chain assisted networks. Common approaches to combatting errors and failures in data storage systems add redundancy to blocks of user data before they are stored. Existing mathematical solutions that describe these operations intrinsically assume that the stored data block can be accessed in its entirety and in a centralized manner. To meet the requirements of emerging systems that have both new types of data tasks and a decentralized data organization, there is now an identified need to develop new mathematical models and abstractions that will provide relevant theoretical foundations and practical design principles for decentralized data storage systems. The project addresses this need by introducing the concept of a sampling channel, which serves to mathematically capture the relationship between the user data and its representation that is available for processing. The key feature of the sampling concept, motivated by the properties of modern data systems, is that it assumes neither centralized access nor full-block availability. The project will then develop mathematical tools and techniques that are suitable for sampling channels of different types. In addition to technical and scientific contributions, this project also has a potential to reduce resource consumption and increase robustness to adversarial participants in future information systems. This project will develop a new mathematical framework centered around sampling channels, with the focus on their applications to decentralized data storage systems. Moving beyond the basic and well-studied case of uncontrolled sampling, emerging distributed systems and applications motivate the usage of controlled sampling, which is classified into two levels depending on whether one has the exact control on the sampling or only through its distribution. Channel coding is a scientific discipline aimed at maximizing the user-data robustness and minimizing system redundancy through principled mathematical frameworks. In this context, the investigators will focus their attention on the theory and practice of efficiently decodable codes with sparse graphical representations. Such codes offer sufficient flexibility to be utilized for the sampling categories and system tasks that are needed in emerging systems. Equipped with initial compelling findings that point to code-design principles that depart from the methods proven successful in conventional settings, the investigators will rigorously analyze existing code families, and offer techniques on how to adapt them to the studied setting, as well as develop new code families. The project will involve theoretically establishing relevant code properties, performance guarantees and trade-offs, and practical evaluations and comparisons. 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.