Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models

Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models

Bibliographic Details
Main Author: Zeng, Yun
Language:English
Published: The Ohio State University / OhioLINK 2018
Subjects:
Industrial Engineering
parallel processing
distributed processing
fork and join queueing networks
blocking
throughput
scalability
Online Access:http://rave.ohiolink.edu/etdc/view?acc_num=osu1524152166819334
id ndltd-OhioLink-oai-etd.ohiolink.edu-osu1524152166819334
record_format oai_dc
collection NDLTD
language English
sources NDLTD
topic Industrial Engineering
parallel processing
distributed processing
fork and join queueing networks
blocking
throughput
scalability
spellingShingle Industrial Engineering
parallel processing
distributed processing
fork and join queueing networks
blocking
throughput
scalability
Zeng, Yun
Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
author Zeng, Yun
author_facet Zeng, Yun
author_sort Zeng, Yun
title Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
title_short Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
title_full Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
title_fullStr Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
title_full_unstemmed Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models
title_sort scalability analysis of parallel and distributed processing systems via fork and join queueing network models
publisher The Ohio State University / OhioLINK
publishDate 2018
url http://rave.ohiolink.edu/etdc/view?acc_num=osu1524152166819334
work_keys_str_mv AT zengyun scalabilityanalysisofparallelanddistributedprocessingsystemsviaforkandjoinqueueingnetworkmodels
_version_ 1719453741637173248
spelling ndltd-OhioLink-oai-etd.ohiolink.edu-osu15241521668193342021-08-03T07:06:26Z Scalability Analysis of Parallel and Distributed Processing Systems via Fork and Join Queueing Network Models Zeng, Yun Industrial Engineering parallel processing distributed processing fork and join queueing networks blocking throughput scalability Parallel and distributed processing systems are foundational to the success of cloud computing and big data analytics. As technology advances, future systems are expected to reach unprecedented scales. A critical issue concerns throughput scalability: whether the throughput performance can be sustained as the systems scale in size. The problem is non-trivial since the processing times are stochastic in nature and the systems are subject to various synchronization constraints and resource capacity limitations. While practical engineering efforts have long been made to build and scale parallel and distributed processing systems, the mathematical foundations toward understanding the throughput performance of ever-growing systems remain rudimentary. In addition, as revealed by Moore's law, technology advances render larger storage space and faster processing speed. To what extent these advancements on resource capabilities can help improve throughput scalability of parallel and distributed processing systems remains unclear.In this dissertation, we investigate the scalability of parallel and distributed processing systems via mathematical modeling. We model parallel and distributed processing systems as Fork-and-Join Queueing Networks with Blocking (FJQN/Bs) and study their throughput performance at extreme scales. The networks can have arbitrary topology, arbitrary heterogeneous storage and processing capabilities, arbitrary initial state, and generally distributed service times. We utilize an infinite sequence of FJQN/Bs to model the growth of the systems in size. The sequence is said to be throughput scalable if the limit infimum of the network throughput is strictly positive as the network size grows to infinity along the sequence. Our objective is to identify critical conditions under which a sequence of FJQN/Bs is throughput scalable. We study the problem under light-tailed service times, under heavy-tailed service times, and under improving resource capabilities. Our main contributions are three-fold:•We propose necessary and sufficient conditions on throughput scalability of FJQN/Bs under light-tailed service times. We introduce a novel topological concept, called minimum level, that plays an important role in determining the scalability of FJQN/Bs. We construct throughput bounds as functions of minimum level, network degree, buffer sizes, and service time distributions.•Under heavy-tailed service times, we propose necessary conditions and sufficient conditions on throughput scalability of FJQN/Bs. We introduce two novel geometrical concepts: scaling dimension and extended metric dimension, and we establish strong connections among the two dimensions, heavy-tailed index of service time distributions, and throughput limits. •We extends our analyses to situations where resource capabilities such as buffer sizes and processing speed are also improving as networks expand in size. Corresponding necessary and/or sufficient conditions on scalability are proposed. We show that appropriate growth of resource capabilities can help mitigate throughput degradation and recover scalability.Our results provide analytical insights and engineering principles for designing parallel and distributed processing systems for next-generation big data analytics. The results can also be used to study other types of scaling networks or fractals such as social networks, electrical grid, Internet of Things, etc. Other industries, such as manufacturing, supply chain, digital communication, traffic, transportation, etc., can also benefit from our results. 2018-08-14 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1524152166819334 http://rave.ohiolink.edu/etdc/view?acc_num=osu1524152166819334 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.

Similar Items