Ensuring the reliability and availability of complex networked services demands effective root cause analysis (RCA) across cloud environments, data centers, and on-premises networks. Traditional RCA methods, which involve manual inspection of data sources such as logs and telemetry data, are often time-consuming and challenging for on-call engineers. While statistical inference methods have been employed to estimate the causality of network events, these approaches alone are similarly challenging and suffer from a lack of interpretability, making it difficult for engineers to understand the predictions made by black-box models. In this paper, we present RCACopilot, an advanced on-call system that combines statistical tests and large language model (LLM) reasoning to automate RCA across various network environments. RCACopilot gathers and synthesizes critical runtime diagnostic information, predicts the root cause of incidents, provides a clear explanatory narrative, and offers targeted action steps for engineers to resolve the issues. By utilizing LLM reasoning techniques and retrieval, RCACopilot delivers accurate and practical support for operators.

We present a conversational AI framework for monitoring and analysis of distributed AI/ML infrastructure based on a multi-agent architecture. The system incorporates specialized agents for topology discovery, health assessment, network flow analysis, and root cause analysis (RCA), accessible through a natural language interface. An agent orchestration component parses and routes user queries, allowing the platform to support a range of cluster observability and troubleshooting tasks using both sequential and parallel agent workflows. Each agent interacts with dedicated analytical microservices via Model Context Protocol (MCP), enabling modular and extensible evaluation of infrastructure state. We detail the system architecture, agent design, setup used for experiments and discuss the implications of conversational AI agents for automated RCA and operational efficiency in monitoring and troubleshooting large-scale ML infrastructure.

We present a log analytics framework that collects collective communication logs from real-time distributed AI/ML workloads such as Large Language Models training, finetuning and inferencing jobs. It is designed to handle massive log volumes from frameworks such as NVIDIA’s Collective Communication Library (NCCL) during distributed GPU operations, the framework ingests heterogeneous log streams as these logs are often the first indicators of faults in the end-to-end path. Using NLP-based parsing, log template clustering, temporal correlation for related logs, and anomaly pattern detection, it links critical log events to performance metrics across layers which enable fault localization in GPU-to-GPU communication. Beyond fault detection, it provides comprehensive performance analytics from byte exchange patterns to identify network bottlenecks like low-latency links, bandwidth-constrained node pairs, and network pressure points that increasingly constrain modern AI workloads. In case of distributed inferencing workloads (ex. DeepSeek R1 Distilled 8B), it captures phase-specific (prefill/decode) and collective operation-specific metrics, revealing communication bottlenecks affecting end-to-end response time.

As sensor nodes are subject to strict power limitations and often deployed in harsh environments, there is significant potential for a node to be hidden from another node. The impact of hidden nodes on performance of the IEEE 802.15.4 low-rate wireless personal area network protocol is evaluated. At lower Tx power, as a result of hidden nodes, there are more collisions in the network increasing the cost of packet delivery. The simulation results indicate that there exists an optimum transmission power that minimizes the impact of hidden nodes, and it depends on factors such as network load, desired throughput and beacon order. Results indicate that additional throughput gain and lower cost per packet delivered can be achieved by increasing the beacon order at higher Tx power.

In many emerging sensing applications, random network losses may lead to drop of critical information, rendering partially received data useless for the end users. A packet-marking scheme based on the application content is proposed that enables application-aware processing of the data within the overlay network. A token-bucket based rate control algorithm in conjunction with the proposed packet-marking scheme enables on-the-fly selection of data for forwarding/drop to a particular end user at the desired transmission rate. We demonstrate the effectiveness of the packet-marking and token-bucket based rate control scheme in simultaneously meeting heterogeneous QoS requirements of the multiple end users for content quality and bandwidth for a weather monitoring sensing application.

A formal approach for characterizing, evaluating and modeling packet reordering is presented. Reordering is, a phenomenon that is likely to become increasingly common on the Internet, and it can have an adverse impact on end-to-end performance and network resource utilization. Reorder density (RD) is a comprehensive metric for reordering. Several properties of RD are presented that provide insight into the nature of reordering present in a sequence. Based on measurements of packet reordering, certain basic patterns of reordering that are prevalent on the Internet are identified. By focusing on these basic patterns, a model is developed for reordering in networks involving two parallel paths, using a simple load-balancing scenario as an example. The model is verified using an emulation testbed

Summary form only given. An ad hoc grid is a heterogeneous computing and communication system without a fixed infrastructure; all of its components are mobile. Energy management is a major concern in an ad hoc grid. One important aspect of energy management is to minimize the energy consumption during a mission. In an ad hoc grid, communication and computations are deeply intertwined, and any energy optimization must consider both types of activities together rather than separately. The mapping (defined as matching and scheduling) of tasks onto machines with varied computational capabilities has been shown, in general, to be an NP-complete problem. Therefore, heuristic techniques are required to efficiently map tasks to machines in an ad hoc grid so as to minimize the energy consumed due to communication and computation. This research evaluates and compares energy management issues for resource allocation in ad hoc grids using six static heuristics.

Out of order arrival of packets is an inevitable phenomenon on the Internet. Application performance can degrade to a great extent due to out-of-order arrival of packets. Metrics to characterize the degree of re-ordering will promote the evaluation of network protocols with respect to packet reordering as well as provide a uniformly defined quantitative measure for packet reordering. We define the concept of "reorder density function" (RD), which quantitatively measures the degree of reordering in sequences of packets. While the reorder density function can provide a comprehensive picture on reordering, measures such as mean and median of reorder density can be used when simpler metrics are desired. Results of reordering analysis using the above metrics for TCP and UDP based applications under different network conditions are presented.