Files in this item

FilesDescriptionFormat

application/pdf

application/pdfSAEED-THESIS-2016.pdf (4MB)
(no description provided)PDF

Description

Title:Sandooq: improving the communication cost and service latency for a multi-user erasure-coded geo-distributed cloud environment
Author(s):Saeed, Shayan
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):erasure-coding
geo-distributed data centers
latency
communication cost
Abstract:Modern data centers have to accommodate the storage of an increasing amount of data with multiple users accessing that data from all over the world. Most of these data centers are geo-distributed to improve availability and protect against the loss of data in the case of outages and disasters. They are also increasingly using erasure codes to improve the reliability at a much lower storage cost. In addition to reliability, the clients and applications also demand storage solutions with better performance and cost-effectiveness. For a geo-distributed data center, a major part of the cost is associated with sending the data between the data centers. This paper builds on previous work to minimize the latency and cost in a data center and applies it to a multi-user geo-distributed environment. We develop a mathematical model for service latency and communication cost for a multi-user geo-distributed cloud environment. We also provide an algorithm to jointly optimize the service latency and communication cost by controlling the placement of the erasure-coded file chunks and scheduling the requests for these chunks. Through simulations, we show that our algorithm converges quickly and outperforms other heuristics in optimizing service latency and communication cost.
Issue Date:2016-04-25
Type:Thesis
URI:http://hdl.handle.net/2142/90633
Rights Information:Copyright 2016 Shayan Saeed
Date Available in IDEALS:2016-07-07
Date Deposited:2016-05


This item appears in the following Collection(s)

Item Statistics