Novel latency bounds for distributed coded storage

Parag, P and Chamberland, J-F (2018) Novel latency bounds for distributed coded storage. In: Information Theory and Applications Workshop, ITA 2018, 11 - 16 February 2018, San Diego.

PDF IEEE_ITA_2018.pdf - Published Version Restricted to Registered users only Download (333kB) | Request a copy

Abstract

Distributed storage systems are rapidly emerging as a desirable paradigm for cloud infrastructures. Through redundancy, such systems can improve the reliability of data backends. Interestingly, distributed storage can also enhance file access times for incoming requests, as it takes advantage of statistical averaging. Still, conducting performance characterization for ensuing delay profiles remains a challenge. Several recent contributions in this area construct bounds on the performance of redundant systems. These are then used to explore the latency-redundancy tradeoff, and assess the relative values of candidate implementations. Along these lines, this work establishes novel upper and lower bounds on the mean sojourn time of a request entering a distributed storage system. These bounds are based on stationary distributions of dominating quasi-birth-death processes and, in many settings, they can be made progressively tighter at the expense of additional computations.

Item Type:	Conference Paper
Publication:	2018 Information Theory and Applications Workshop, ITA 2018
Publisher:	Institute of Electrical and Electronics Engineers Inc.
Additional Information:	The copyright for this article belongs to the IEEE.
Keywords:	Information theory; Multiprocessing systems; Redundancy, Cloud infrastructures; Conducting performance; Distributed storage; Distributed storage system; Quasi birth-death process; Stationary distribution; Statistical Averaging; Upper and lower bounds, Digital storage
Department/Centre:	Division of Electrical Sciences > Electrical Communication Engineering
Date Deposited:	03 Aug 2022 06:23
Last Modified:	03 Aug 2022 06:23
URI:	https://eprints.iisc.ac.in/id/eprint/75199

Actions (login required)

View Item