Resource management of replicated service systems provisioned in the cloud

Abstract

I certify that except where due acknowledgement has been given, the work presented in this thesis is that of the author alone; the work has not been submitted previously, in whole or in part, to qualify for any other academic award; and the content of the thesis is the result of work which has been carried out since the official commencement date of the approved research program. Lugano, 11 February 2015 ×ØÖ Ø Service providers seek scalable and cost-effective cloud solutions for hosting their applications. Despite significant recent advances facilitating the deployment and management of services on cloud platforms, a number of challenges still remain. Service providers are confronted with time-varying requests for the provided applications, inter-dependencies between different components, performance variability of the procured virtual resources, and cost structures that differ from conventional data centers. Moreover, fulfilling service level agreements, such as the throughput and response time percentiles, becomes of paramount importance for ensuring business advantages. Mathias BjörkqvistIn this thesis, we explore service provisioning in clouds from multiple points of view. The aim is to best provide service replicas in the form of VMs to various service applications, such that their tail throughput and tail response times, as well as resource utilization, meet the service level agreements in the most cost effective manner. In particular, we develop models, algorithms and replication strategies that consider multi-tier composed services provisioned in clouds. We also investigate how a service provider can opportunistically take advantage of observed performance variability in the cloud. Finally, we provide means of guaranteeing tail throughput and response times in the face of performance variability of VMs, using Markov chain modeling and large deviation theory. We employ methods from analytical modeling, event-driven simulations and experiments. Overall, this thesis provides not only a multi-faceted approach to exploring several crucial aspects of hosting services in clouds, i.e., cost, tail throughput, and tail response times, but our proposed resource management strategies are also rigorously validated via trace-driven simulation and extensive experiments.

Cloud Application Predictability through Integrated Load-Balancing and Service Time Control

“…V. RELATED WORK Building distributed systems that offer guarantees on their timely execution while the system is subject to uncertainty and changes is a challenging task. Bounding latencies is of utmost importance, but this is quite difficult in the presence of changes [5,6,14,24,25,42]. Changes are unpredictable, they can be dramatic, and they can include malfunctioning [23], slow down [12], failures [18], and much more.…”

“…To determine the size of data centers, and properly dimension the resources to be allocated in each geographic location, most data center owners use predictions of the computational needs [29,36]. The computational resource within a data center is then used to serve requests coming from multiple clients, providing the illusion of infinite capacity and, as a result, the possibility of bounding the latency [5,6,14,24,25,42]. To do so the architecture uses multiple instances of the same application, here called replicas, and predictions and estimations of traffic and needed computational capacity.…”

“…For example, load-balancing strategies based on response times are to be avoided when a replica control strategy that bounds the response times is used [11]. This is not only true for brownout, but for every technique that enforces bounded response times [5,6], like admission control policies [26,38].…”

Formal foundations of serverless computing

“…Serverless computing and other container-based platforms suffer several performance and utilization barriers. There are several ways to address these problems, including datacenter design [Gan and Delimitrou 2018], resource allocation [Björkqvist et al 2016], programming abstractions [Baldini et al 2017;Rabbah 2017], edge computing [Aske and Zhao 2018], and cloud container design [Shen et al 2019]. λ is designed to elucidate subtle semantic issues (not performance problems) that affect programmers building serverless applications.…”