High performance image storage-retrieval architecture

Abstract

Enterprises today need to manage their content/ documents more efficiently than in the past. Even across different domains the complexity of managing millions of documents is huge. One such case was when our company was invited to submit a proposal for a central bank of a large country for its Cheque Truncation system. This would help the central bank reduce the turn-around time for clearing cheques at the same time save costs in physically transporting, storing and processing these cheques. Our company evaluated two well known approaches by doing Proof of Concept exercises on two architectures which were identified to be proposed to the bank. During the evaluation, strengths and weaknesses of both architectures were quantified on different parameters to identify a more suitable architecture. However both architectures were found to be unsuitable to meet the non-functional requirements of this bank, primarily Performance, Scalability and Reliability. Hence we began studying the problem of storing and retrieving images from a system where millions of images are added per day, with an aim of getting superior performance, scalability and high reliability. This paper describes the two architectures that were evaluated and adduces benchmark numbers to prove their unsuitability against the given requirements. Next it describes a set of experiments that were performed. Based on experimental findings a new architecture was developed. This architecture is capable of delivering extremely high-performance, high-scalability and high-reliability. This architecture is what is described and labeled as High-Performance ImageStore Solution in this paper.