A monitoring system for collecting and aggregating metrics from distributed clouds

Tamara Ranković, Mateja Rilak, Janko Rakonjac, Miloš Simić · 2026

Read A monitoring system for collecting and aggregating metrics from distributed clouds by Tamara Ranković, Mateja Rilak, Janko Rakonjac, Miloš Simić (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-05241.txt.

Introduction

Tamara Ranković and colleagues design and implement a monitoring system collecting and aggregating metrics from distributed clouds, where users dynamically create and dispose of tailored ad-hoc clouds. Observability is essential for the model in practice, so node agents collect machine-level, container-level, and application-level metrics. In health checks, that data moves to the cloud's control plane, where it is persisted and served through several APIs, including a streaming one. Node metrics are also aggregated per cloud, giving clients a more comprehensive view of system state.

Catalog record

CREATORTamara Ranković, Mateja Rilak, Janko Rakonjac, Miloš Simić
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24486
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.05241

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A monitoring system for collecting and aggregating metrics from distributed clouds Thanks: Funded by the European Union (TaRDIS, 101093006). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them. Tamara Ranković, Mateja Rilak, Janko Rakonjac, Miloš Simić Abstract Applications requiring real-time processing of large volumes of data have been the main driver for rethinking the traditional cloud, giving rise to novel cloud models. Distributed cloud (DC) is a model that allows users to dynamically create and dispose of strategically located ad-hoc clouds that contain resources best tailored to their needs. It is essential for this model to provide a high degree of observability for it to be viable in real-world scenarios. In this paper, we present the design and implementation of a monitoring system that collects metrics from DCs and makes them accessible to diverse clients. Agents running on nodes are responsible for collecting machine-, container-, and application-level metrics. During the health-check protocol, that data is transferred from the node to DC’s control plane running inside the cloud. There, it is persisted and served via multiple APIs, including a streaming API. Moreover, node metrics are aggregated for every DC in order to provide a more comprehensive view of the system’s state. I Introduction In the past decade, we have seen a pronounced transformation of the cloud model. Requirements posed by modern applications have been the main driver of this transformation. Real-time transfer, processing, and retrieval of large volumes of data have been shown to be inefficient or impossible when relying on remote centralized datacenters [ 4 ] . To address these issues, many novel cloud models, like edge and fog computing, have emerged [ 17 , 1 ] . While proposing different solutions, they all embrace the concept of data locality - processing data on locations where it is stored, but also storing it near locations where it is generated. Modern manufacturing facilities, power plants, autonomous vehicles, and other related use cases, where data sources and consumers are in near proximity and real-time data processing and retrieval are essential, benefit from this organization tremendously. Besides introducing efficiency, data privacy and security are also enhanced, as data may never cross the boundary of trusted locations or networks. The distributed cloud (DC) model is one of the models concerned with presented problems [ 20 , 21 ] . It consists of cloud and DC layers. The cloud layer is the traditional centralized cloud, while the DC layer is meant to offload the cloud layer and also serve clients with strict latency and privacy requirements. The DC layer consists of dynamic DCs formed and discarded by users. Each DC should be tailored to the user’s needs, including its resource capacity and location [ 20 , 6 ] . The lifecycle of a DC is coordinated by a control plane in the cloud layer, but the DC itself ensures its proper operation, providing a hybrid architecture. Users are expected to specify configuration and workloads for their DCs, while the system is responsible for its convergence to the desired state. This model allows for seamless cooperation between the cloud and the DC. To make the DC model a viable option for real-world applications, we need to provide a rich ecosystem of functionalities and reliable tools around it. Efforts towards this have been made by introducing namespaces and dataspaces, allowing for virtual DCs that provide multitenancy capabilities [ 18 , 19 ] . Moreover, mechanisms for dynamic system and application reconfiguration were developed [ 14 ] , with native support for misconfiguration prevention and detection [ 15 ] . Access to any resource in the DC is managed by a hierarchical attribute-based model [ 16 , 13 ] . Considering DC’s complex architecture and interactions among its components, it is necessary to ensure a high degree of observability. To achieve this in any environment, it is a standard practice to collect various metrics, logs, and traces [ 9 ] . They serve as a foundation for monitoring the system’s behavior and performance and detecting any bottlenecks or failures. Moreover, workload schedulers and infrastructure autoscalers rely primarily on metrics as inputs for making decisions. Existing monitoring solutions have become highly sophisticated and have numerous capabilities, however, none of them are native to the DC model. As DCs are dynamic, regularly being formed or destroyed, and with many nodes constantly joining or leaving, the monitoring system needs to be flexible enough to accommodate these fluctuations. Additionally, as hardware comprising DCs can often be resource-limited, lightweight tools are essential to ensure minimal resource consumption by system components. The focus of this paper is on designing and implementing a monitoring system native for DCs, more precisely for collecting metrics from DCs and making them accessible to clients. We consider end users who manage DCs the primary clients, as it is crucial for them to have insight into the state of their DC infrastructure and applications running on it. However, we take more potential clients into consideration, including workload schedulers, infrastructure, and application-level autoscalers, who make decisions based on this data. Moreover, if these clients incorporate machine learning models, they can utilize collected metrics also as training sets. In Section II we discuss existing monitoring solutions, particularly the cloud-native ones. Sections III and IV present our solution, its design and implementation respectively. In Section V we provide reasoning behind certain design decisions and discuss the system’s limitations and possible improvements, while Section VI concludes the paper and outlines directions for future work. II Related work Prometheus 1 1 1 Prometheus: https://prometheus.io/ is a widely used monitoring system with a rich set of tools and integrations. It is a centralized metrics collection and analysis solution that uses a pull communication model. A Prometheus instance is configured to periodically scrape metrics from desired targets and store them in its time-series database. Targets can expose their metrics directly when instrumented with client libraries. However, when that is not possible, usually in third-party systems, numerous exporters have been developed to extract and transform metrics from such systems into the Prometheus format. The standardization of this format, called OpenMetrics 2 2 2 OpenMetrics: https://openmetrics.io/ , was introduced to encourage interoperability between different monitoring tools. Kubernetes 3 3 3 Kubernetes: https://kubernetes.io/ , a popular container orchestrator, has a resource metrics pipeline and a full metrics pipeline. The first one serves a limited set of metrics via the metrics-server which collects metrics from nodes and stores them only in memory for a short time. This pipeline is primarily meant to be used by autoscalers. The second pipeline provides a significantly richer set of metrics from nodes but requires users to implement their adapter that exposes those metrics. All metrics are transmitted in the OpenMetrics format. When orchestrating workloads over a federation of clusters, a scheduler needs to have a global view of the entire federation. Acala [ 7 ] is designed to aggregate metrics from federations of Kubernetes clusters. Global View Cluster contains Prometheus storage and a set of Acala controllers that periodically request metrics from clusters. There is an Acala member in

Read the complete text