<p>Microservices have emerged as a popular pattern for developing large-scale applications in cloud environments for their flexibility, scalability, and agility benefits. However, microservices make management more complex due to their scale, multiple languages, and distributed nature. Orchestration and automation tools like Kubernetes help deploy microservices running simultaneously, but it can be difficult for an operator to understand their behaviors, interdependencies, and interactions. In such a complex and dynamic environment, performance problems (e.g., slow application responses and high resource usage)  require significant human effort spent on diagnosis and recovery. Moreover, manual diagnosis of cloud microservices tends to be tedious, time-consuming, and impractical. Effective and automated performance analysis and anomaly detection require an observable system, which means an application's internal state can be inferred by observing and tracking metrics, traces, and logs. Traditional APM uses libraries and SDKs to improve application monitoring and tracing but has additional overheads of rewriting, recompiling, and redeploying the applications' code base. Therefore, there is a critical need for a standardized automated microservices observability solution that does not require rewriting or redeploying the application to keep up with the agility of microservices.</p> <p><br></p> <p>This thesis studies observability for microservices and implements an automated Extended Berkeley Packet Filter (eBPF) based observability solution. eBPF is a Linux feature that allows us to write extensions to the Linux kernel for security and observability use cases. eBPF does not require modifying the application layer and instrumenting the individual microservices. Instead, it instruments the kernel-level API calls, which are common across all hosts in the cluster. eBPF programs provide observability information from the lowest-level system calls and can export data without additional performance overhead. The Prometheus time-series database is leveraged to store all the captured metrics and traces for analysis. With the help of our tool, a DevOps engineer can easily identify abnormal behavior of microservices and enforce appropriate countermeasures. Using Chaos Mesh, we inject anomalies at the network and host layer, which we can identify with root cause identification using the proposed solution. The Chameleon cloud testbed is used to deploy our solution and test its capabilities and limitations.</p>

<p>Physical security is one of the most basic human needs. People care about it for various reasons; for the safety and security of personnel, to protect private assets, to prevent crime, and so forth. With the recent proliferation of AI, various smart physical security systems are getting introduced to the world. Many researchers and engineers are working on developing AI-driven physical security systems that have the capability to identify potential security threats by monitoring and analyzing data collected from various sensors. One of the most popular ways to detect unauthorized entrance to restricted space is using face recognition. With a collected stream of images and a proper algorithm, security systems can recognize faces detected from the image and send an alert when unauthorized faces are recognized. In recent years, there has been active research and development on neural networks for face recognition, e.g. FaceNet is one of the advanced algorithms. However, not much work has been done to showcase what kind of end-to-end system architecture is effective for running heavy-weight computational loads such as neural network inferences. Thus, this study explores different hardware options that can be used in security systems powered by a state-of-the-art face recognition algorithm and proposes that an edge computing based approach can significantly reduce the overall system latency and enhance the system reactiveness. To analyze the pros and cons of the proposed system, this study presents two different end-to-end system architectures. The first system is an edge computing-based system that operates most of the computational tasks at the edge node of the system, and the other is a traditional application server-based system that performs core computational tasks at the application server. Both systems adopt domain-specific hardware, Tensor Processing Units, to accelerate neural network inference. This paper walks through the implementation details of each system and explores its effectiveness. It provides a performance analysis of each system with regard to accuracy and latency and outlines the pros and cons of each system.</p> <p><br></p>

<p><em>Software Instrumentation is the process of collecting data during an application’s runtime,</em></p> <p><em>which will help us debug, detect errors and optimize the performance of the binary. The</em></p> <p><em>recent increase in demand for low latency and high throughput systems has introduced new</em></p> <p><em>challenges to the process of Software Instrumentation. Software Instrumentation, especially</em></p> <p><em>dynamic, has a huge impact on systems performance in scenarios where there is no early</em></p> <p><em>knowledge of data to be collected. Naive approaches collect too much or too little</em></p> <p><em>data, negatively impacting the system’s performance.</em></p> <p><em>This thesis investigates the overhead added by reservoir sampling algorithms at different</em></p> <p><em>levels of granularity in real-time instrumentation of distributed software systems. Also, this thesis describes the implementation of sampling techniques and algorithms to reduce the overhead caused by instrumentation.</em></p>