Improving interoperability in distributed multi-tier software stacks

Ramsingh, Adrian (2023) Improving interoperability in distributed multi-tier software stacks. PhD thesis, University of Glasgow.

Abstract

Distributed multi-tier software stacks organise and deploy software components as a hierarchy of interacting tiers. The components are typically heterogeneous, i.e. each component may be written in a different language and may interoperate using a variety of protocols. Tiered software is modular but leads to a range of interoperability challenges including the following. (1) Interoperating components in multiple languages and paradigms increases developer cognitive load since they must simultaneously reason in multiple languages and paradigms. (2) There must be correct interoperation of components, e.g. adherence to the API or communication protocols between components. (3) Interoperation between different components can lead to diverse modes of failure as each component can fail in unique ways. Many of these challenges are the result of contributing factors like tight coupling or polyglot programmming.
This thesis investigates techniques to improve heterogeneous interoperability in distributed multi-tier software stacks. Some common approaches include microservices and tierless languages.
Microservices are perceived to offer better reliability than components in multi-tier software stacks through the loose coupling of services. The reliability of microservices is investigated by combining the established properties of dependence and state with reliability. This defines a new three-dimensional space: the Microservices Dependency State Reliability (MDSR) classification with six classes. The feasibility of statically identifying MDSR classes is demonstrated with a prototype analyser that identifies all six classes in Flask microservices web applications. The reliability implications of the different MDSR classes are evaluated by running three case study applications (Hipster-Shop, JPyL & WordPress) against a fault injector. Key results are as follows. (1) All applications fail catastrophically if a critical microservice fails. (2) Applications survive the failure of individual minor microservice(s). (3) The failure of any chain of microservices in JPyL & Hipster is catastrophic. (4) Individual microservices do not necessarily have minor reliability implications.
In a tierless language, the compiler generates the code for each component and ensures their correct interoperation. They are mainly used to implement web stacks. However, their use in implementing IoT stacks is less common. This investigation compares interoperation in tiered and tierless IoT stacks through the systematic evaluation of four implementations of the prototype UoG smart campus IoT system: two tierless and two Python-based tiered. Key results of the study are as follows. (1) Tierless languages have the potential to significantly reduce the development effort for IoT systems, requiring 70% less code than the tiered implementations. (2) Tierless languages have the potential to significantly improve the reliability of IoT systems. (3) The first comparison of a tierless codebase for resource-rich sensor nodes and one for resourceconstrained sensor nodes shows that they have very similar functional structure and code sizes - within 7%.
Tier elimination is a technique that removes a tier/component by integrating two tiers. Specifically, this thesis investigates the implications of eliminating the Apache web server in a 4-tier web stack: Jupyter Notebook, Apache, Python, Linux (JAPyL) and replacing it with PHP libraries in the frontend webpage to get the 3-tier (JPL). The study reveals the following. (1) The JPL 3-tier web stack requires that the developer uses fewer programming languages and paradigms than JAPyL, i.e two compared with four languages and two compared with three paradigms. (2) JPL requires 42% less code than JAPyL. (3) In JPL, some of the functionalities can be automated due to the decreased abstraction levels at the upper layers of the stack. (4) However, the latency in JPL is two to three times greater than that of JAPyL. So while tier elimination reduces developer effort and semantic friction the tradeoffs are high performance overhead & resource consumption and increasing code complexity.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: Q Science > QA Mathematics > QA76 Computer software
• Colleges/Schools: College of Science and Engineering > School of Computing Science
• Funder's Name: Engineering and Physical Sciences Research Council (EPSRC), Engineering and Physical Sciences Research Council (EPSRC)
• Supervisor's Name: Singer, Dr. Jeremy and Trinder, Professor Phil
• Date of Award: 2023
• Depositing User: Theses Team
• Unique ID: glathesis:2023-83372
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 20 Jan 2023 09:45
• Last Modified: 20 Jan 2023 09:48
• Thesis DOI: 10.5525/gla.thesis.83372
• URI: https://theses.gla.ac.uk/id/eprint/83372

Related URLs

• Article DOI
• Enlighten Publications Record
• Article DOI
• Enlighten Publications Record
• Article DOI
• Enlighten Publications Record