arXiv subjects

Francesco Bronzino

Publications and source records attributed to Francesco Bronzino.

At least 19 recordsLinked to original sources

arXiv2026

ChronoSSM: Training for Temporally Aware Representations in Autoregressive State Space Models

Modern sequence models, from Transformers to State Space Models, have enabled powerful generative modeling across diverse domains, yet they are typically trained to predict what happens while treating when it happens as a secondary concern. In data-mining settings where events are associated with explicit timing information, this separation can limit temporal reasoning, anomaly detection, and faithful reconstruction of event chronology. A common strategy is to treat timing as an auxiliary signal, training a separate timing model using representations learned solely for event prediction. However, this two-stage approach implicitly assumes that representations optimized for event prediction already contain sufficient temporal structure. We introduce ChronoSSM, an autoregressive State Space Model (SSM) that jointly models events and timestamps with a shared backbone trained using combined token and temporal generation objectives. We compare the joint regime, where temporal supervision updates the backbone, with the two-stage regime, where timing is learned only using the frozen event representations. Across four domains spanning dense and partial timestamp supervision, joint training consistently makes inter-arrival information more recoverable from frozen representations without any systematic degradation in content-generation quality overall. Our results show that temporal supervision can produce more temporally informative representations without materially degrading autoregressive event modeling.

cs.LG↗

arXiv2026

Roomie: Interference-Aware Colocation for Efficient Model Serving

As demand for DNN inference grows, GPU capacity is increasingly oversubscribed, forcing operators to colocate multiple models on the same device in both cloud and edge deployments. Whether colocation succeeds or violates SLOs depends on the temporal overlap of kernels from concurrently executing models -- an effect that existing serving systems either ignore or approximate using aggregate resource profiles that fail to capture temporal dynamics. This paper presents Roomie, a model serving orchestration architecture that predicts and avoids kernel-level interference between colocated DNNs. Roomie decouples offline kernel profiling from online interference prediction. It uses profiling only to extract per-kernel resource configurations, and predicts interference with an occupancy-based analytical model immune to profiler-induced timing distortion. A pairwise greedy heuristic then approximates multi-model interference in polynomial rather than exponential time, and an online placement algorithm then uses these estimates to assign each incoming model to the GPU that minimizes predicted slowdown. Our experimental evaluation compares Roomie against state-of-the-art solutions across both cloud-grade server clusters and embedded edge devices, demonstrating that Roomie reduces SLO violations (i.e., inference latency) by up to 3x, while maintaining comparable, and in many cases superior, goodput relative to existing approaches.

cs.DC↗

arXiv2026

Characterizing the Impact of Active Queue Management on Speed Test Measurements

Present day speed test tools measure peak throughput, but often fail to capture the user-perceived responsiveness of a network connection under load. Recently, platforms such as NDT, Ookla Speedtest and Cloudflare Speed Test have introduced metrics such as ``latency under load'' or ``working latency'' to fill this gap. Yet, the sensitivity of these metrics to basic network configurations such as Active Queue Management (AQM) remains poorly understood. In this work, we conduct an empirical study of the impact of AQM on speed test measurements in a laboratory setting. Using controlled experiments, we compare the distribution of throughput and latency under different load measurements across different AQM schemes, including CoDel, FQ-CoDel and Stochastic Fair Queuing (SFQ). On comparing with a standard drop-tail baseline, we find that measurements have high variance across AQM schemes and load conditions. These results highlight the critical role of AQM in shaping how emerging latency metrics should be interpreted, and underscore the need for careful calibration of speed test platforms before their results are used to guide policy or regulatory outcomes.

cs.NI↗

arXiv2026

Less is More: Optimizing Probe Selection Using Shared Latency Anomalies

Latency anomalies, defined as persistent or transient increases in round-trip time (RTT), are common in residential Internet performance. When multiple users observe anomalies to the same destination, this may reflect shared infrastructure, routing behavior, or congestion. Inferring such shared behavior is challenging because anomaly magnitudes vary widely across devices, even within the same ISP and geographic area, and detailed network topology information is often unavailable. We study whether devices experiencing a shared latency anomaly observe similar changes in RTT magnitude using a topology-agnostic approach. Using four months of high-frequency RTT measurements from 99 residential probes in Chicago, we detect shared anomalies and analyze their consistency in amplitude and duration without relying on traceroutes or explicit path information. Building on prior change-point detection techniques, we find that many shared anomalies exhibit similar amplitude across users, particularly within the same ISP. Motivated by this observation, we design a sampling algorithm that reduces redundancy by selecting representative devices under user-defined constraints. Our approach captures 95 percent of aggregate anomaly impact using fewer than half of the deployed probes. Compared to two baselines, it identifies significantly more unique anomalies at comparable coverage levels. We further show that geographic diversity remains important when selecting probes within a single ISP, even at city scale. Overall, our results demonstrate that anomaly amplitude and duration provide effective topology-independent signals for scalable monitoring, troubleshooting, and cost-efficient sampling in residential Internet measurement.

cs.NI↗

arXiv2026

NetSSM: Multi-Flow and State-Aware Network Trace Generation using State Space Models

Access to raw network traffic data is essential for many computer networking tasks, from traffic modeling to performance evaluation. Unfortunately, this data is scarce due to high collection costs and governance rules. Previous efforts explore this challenge by generating synthetic network data, but fail to reliably handle multi-flow sessions, struggle to reason about stateful communication in moderate to long-duration network sessions, and lack robust evaluations tied to real-world utility. We propose a new method based on state space models called NetSSM that generates raw network traffic at the packet-level granularity. Our approach captures interactions between multiple, interleaved flows -- an objective unexplored in prior work -- and effectively reasons about flow-state in sessions to capture traffic characteristics. NetSSM accomplishes this by learning from and producing traces 8x and 78x longer than existing transformer-based approaches. Evaluation results show that our method generates high-fidelity traces that outperform prior efforts in existing benchmarks. We also find that NetSSM's traces have high semantic similarity to real network data regarding compliance with standard protocol requirements and flow and session-level traffic characteristics.

cs.NI↗

arXiv2025

Cruise Control: Dynamic Model Selection for ML-Based Network Traffic Analysis

Modern networks increasingly rely on machine learning models for real-time insights, including traffic classification, application quality of experience inference, and intrusion detection. However, existing approaches prioritize prediction accuracy without considering deployment constraints or the dynamism of network traffic, leading to potentially suboptimal performance. Because of this, deploying ML models in real-world networks with tight performance constraints remains an open challenge. In contrast with existing work that aims to select an optimal candidate model for each task based on offline information, we propose an online, system-driven approach to dynamically select the best ML model for network traffic analysis. To this end, we present Cruise Control, a system that pre-trains several models for a given task with different accuracy-cost tradeoffs and selects the most appropriate model based on lightweight signals representing the system's current traffic processing ability. Experimental results using two real-world traffic analysis tasks demonstrate Cruise Control's effectiveness in adapting to changing network conditions. Our evaluation shows that Cruise Control improves median accuracy by 2.78% while reducing packet loss by a factor of four compared to offline-selected models.

cs.NI↗

arXiv2025

CATO: End-to-End Optimization of ML-Based Traffic Analysis Pipelines

Machine learning has shown tremendous potential for improving the capabilities of network traffic analysis applications, often outperforming simpler rule-based heuristics. However, ML-based solutions remain difficult to deploy in practice. Many existing approaches only optimize the predictive performance of their models, overlooking the practical challenges of running them against network traffic in real time. This is especially problematic in the domain of traffic analysis, where the efficiency of the serving pipeline is a critical factor in determining the usability of a model. In this work, we introduce CATO, a framework that addresses this problem by jointly optimizing the predictive performance and the associated systems costs of the serving pipeline. CATO leverages recent advances in multi-objective Bayesian optimization to efficiently identify Pareto-optimal configurations, and automatically compiles end-to-end optimized serving pipelines that can be deployed in real networks. Our evaluations show that compared to popular feature optimization techniques, CATO can provide up to 3600x lower inference latency and 3.7x higher zero-loss throughput while simultaneously achieving better model performance.

cs.NI↗

arXiv2024

Beyond Data Points: Regionalizing Crowdsourced Latency Measurements

Despite significant investments in access network infrastructure, universal access to high-quality Internet connectivity remains a challenge. Policymakers often rely on large-scale, crowdsourced measurement datasets to assess the distribution of access network performance across geographic areas. These decisions typically rest on the assumption that Internet performance is uniformly distributed within predefined social boundaries. However, this assumption may not be valid for two reasons: crowdsourced measurements often exhibit non-uniform sampling densities within geographic areas; and predefined social boundaries may not align with the actual boundaries of Internet infrastructure. In this paper, we present a spatial analysis on crowdsourced datasets for constructing stable boundaries for sampling Internet performance. We hypothesize that greater stability in sampling boundaries will reflect the true nature of Internet performance disparities than misleading patterns observed as a result of data sampling variations. We apply and evaluate a series of statistical techniques to: aggregate Internet performance over geographic regions; overlay interpolated maps with various sampling unit choices; and spatially cluster boundary units to identify contiguous areas with similar performance characteristics. We assess the effectiveness of the techniques we apply by comparing the similarity of the resulting boundaries for monthly samples drawn from the dataset. Our evaluation shows that the combination of techniques we apply achieves higher similarity compared to directly calculating central measures of network metrics over census tracts or neighborhood boundaries. These findings underscore the important role of spatial modeling in accurately assessing and optimizing the distribution of Internet performance, to inform policy, network operations, and long-term planning decisions.

cs.NI↗

arXiv2024

Ironing the Graphs: Toward a Correct Geometric Analysis of Large-Scale Graphs

Graph embedding approaches attempt to project graphs into geometric entities, i.e, manifolds. The idea is that the geometric properties of the projected manifolds are helpful in the inference of graph properties. However, if the choice of the embedding manifold is incorrectly performed, it can lead to incorrect geometric inference. In this paper, we argue that the classical embedding techniques cannot lead to correct geometric interpretation as they miss the curvature at each point, of manifold. We advocate that for doing correct geometric interpretation the embedding of graph should be done over regular constant curvature manifolds. To this end, we present an embedding approach, the discrete Ricci flow graph embedding (dRfge) based on the discrete Ricci flow that adapts the distance between nodes in a graph so that the graph can be embedded onto a constant curvature manifold that is homogeneous and isotropic, i.e., all directions are equivalent and distances comparable, resulting in correct geometric interpretations. A major contribution of this paper is that for the first time, we prove the convergence of discrete Ricci flow to a constant curvature and stable distance metrics over the edges. A drawback of using the discrete Ricci flow is the high computational complexity that prevented its usage in large-scale graph analysis. Another contribution of this paper is a new algorithmic solution that makes it feasible to calculate the Ricci flow for graphs of up to 50k nodes, and beyond. The intuitions behind the discrete Ricci flow make it possible to obtain new insights into the structure of large-scale graphs. We demonstrate this through a case study on analyzing the internet connectivity structure between countries at the BGP level.

cs.CG↗

arXiv2024

Feasibility of State Space Models for Network Traffic Generation

Many problems in computer networking rely on parsing collections of network traces (e.g., traffic prioritization, intrusion detection). Unfortunately, the availability and utility of these collections is limited due to privacy concerns, data staleness, and low representativeness. While methods for generating data to augment collections exist, they often fall short in replicating the quality of real-world traffic In this paper, we i) survey the evolution of traffic simulators/generators and ii) propose the use of state-space models, specifically Mamba, for packet-level, synthetic network trace generation by modeling it as an unsupervised sequence generation problem. Early evaluation shows that state-space models can generate synthetic network traffic with higher statistical similarity to real traffic than the state-of-the-art. Our approach thus has the potential to reliably generate realistic, informative synthetic network traces for downstream tasks.

cs.NI↗

arXiv2023

NetDiffusion: Network Data Augmentation Through Protocol-Constrained Traffic Generation

Datasets of labeled network traces are essential for a multitude of machine learning (ML) tasks in networking, yet their availability is hindered by privacy and maintenance concerns, such as data staleness. To overcome this limitation, synthetic network traces can often augment existing datasets. Unfortunately, current synthetic trace generation methods, which typically produce only aggregated flow statistics or a few selected packet attributes, do not always suffice, especially when model training relies on having features that are only available from packet traces. This shortfall manifests in both insufficient statistical resemblance to real traces and suboptimal performance on ML tasks when employed for data augmentation. In this paper, we apply diffusion models to generate high-resolution synthetic network traffic traces. We present NetDiffusion, a tool that uses a finely-tuned, controlled variant of a Stable Diffusion model to generate synthetic network traffic that is high fidelity and conforms to protocol specifications. Our evaluation demonstrates that packet captures generated from NetDiffusion can achieve higher statistical similarity to real data and improved ML model performance than current state-of-the-art approaches (e.g., GAN-based approaches). Furthermore, our synthetic traces are compatible with common network analysis tools and support a myriad of network tasks, suggesting that NetDiffusion can serve a broader spectrum of network analysis and testing tasks, extending beyond ML-centric applications.

cs.NI↗

arXiv2023

Internet Localization of Multi-Party Relay Users: Inherent Friction Between Internet Services and User Privacy

Internet privacy is increasingly important on the modern Internet. Users are looking to control the trail of data that they leave behind on the systems that they interact with. Multi-Party Relay (MPR) architectures lower the traditional barriers to adoption of privacy enhancing technologies on the Internet. MPRs are unique from legacy architectures in that they are able to offer privacy guarantees without paying significant performance penalties. Apple's iCloud Private Relay is a recently deployed MPR service, creating the potential for widespread consumer adoption of the architecture. However, many current Internet-scale systems are designed based on assumptions that may no longer hold for users of privacy enhancing systems like Private Relay. There are inherent tensions between systems that rely on data about users -- estimated location of a user based on their IP address, for example -- and the trend towards a more private Internet. This work studies a core function that is widely used to control network and application behavior, IP geolocation, in the context of iCloud Private Relay usage. We study the location accuracy of popular IP geolocation services compared against the published location dataset that Apple publicly releases to explicitly aid in geolocating PR users. We characterize geolocation service performance across a number of dimensions, including different countries, IP version, infrastructure provider, and time. Our findings lead us to conclude that existing approaches to IP geolocation (e.g., frequently updated databases) perform inadequately for users of the MPR architecture. For example, we find median location errors >1,000 miles in some countries for IPv4 addresses using IP2Location. Our findings lead us to conclude that new, privacy-focused, techniques for inferring user location may be required as privacy becomes a default user expectation on the Internet.

cs.NI↗

arXiv2023

AC-DC: Adaptive Ensemble Classification for Network Traffic Identification

Accurate and efficient network traffic classification is important for many network management tasks, from traffic prioritization to anomaly detection. Although classifiers using pre-computed flow statistics (e.g., packet sizes, inter-arrival times) can be efficient, they may experience lower accuracy than techniques based on raw traffic, including packet captures. Past work on representation learning-based classifiers applied to network traffic captures has shown to be more accurate, but slower and requiring considerable additional memory resources, due to the substantial costs in feature preprocessing. In this paper, we explore this trade-off and develop the Adaptive Constraint-Driven Classification (AC-DC) framework to efficiently curate a pool of classifiers with different target requirements, aiming to provide comparable classification performance to complex packet-capture classifiers while adapting to varying network traffic load. AC-DC uses an adaptive scheduler that tracks current system memory availability and incoming traffic rates to determine the optimal classifier and batch size to maximize classification performance given memory and processing constraints. Our evaluation shows that AC-DC improves classification performance by more than 100% compared to classifiers that rely on flow statistics alone; compared to the state-of-the-art packet-capture classifiers, AC-DC achieves comparable performance (less than 12.3% lower in F1-Score), but processes traffic over 150x faster.

cs.NI↗

arXiv2023

LEAF: Navigating Concept Drift in Cellular Networks

Operational networks commonly rely on machine learning models for many tasks, including detecting anomalies, inferring application performance, and forecasting demand. Yet, model accuracy can degrade due to concept drift, whereby the relationship between the features and the target to be predicted changes. Mitigating concept drift is an essential part of operationalizing machine learning models in general, but is of particular importance in networking's highly dynamic deployment environments. In this paper, we first characterize concept drift in a large cellular network for a major metropolitan area in the United States. We find that concept drift occurs across many important key performance indicators (KPIs), independently of the model, training set size, and time interval -- thus necessitating practical approaches to detect, explain, and mitigate it. We then show that frequent model retraining with newly available data is not sufficient to mitigate concept drift, and can even degrade model accuracy further. Finally, we develop a new methodology for concept drift mitigation, Local Error Approximation of Features (LEAF). LEAF works by detecting drift; explaining the features and time intervals that contribute the most to drift; and mitigates it using forgetting and over-sampling. We evaluate LEAF against industry-standard mitigation approaches (notably, periodic retraining) with more than four years of cellular KPI data. Our initial tests with a major cellular provider in the US show that LEAF consistently outperforms periodic and triggered retraining on complex, real-world data while reducing costly retraining operations.

cs.NI↗

arXiv2021

Traffic Refinery: Cost-Aware Data Representation for Machine Learning on Network Traffic

Network management often relies on machine learning to make predictions about performance and security from network traffic. Often, the representation of the traffic is as important as the choice of the model. The features that the model relies on, and the representation of those features, ultimately determine model accuracy, as well as where and whether the model can be deployed in practice. Thus, the design and evaluation of these models ultimately requires understanding not only model accuracy but also the systems costs associated with deploying the model in an operational network. Towards this goal, this paper develops a new framework and system that enables a joint evaluation of both the conventional notions of machine learning performance (e.g., model accuracy) and the systems-level costs of different representations of network traffic. We highlight these two dimensions for two practical network management tasks, video streaming quality inference and malware detection, to demonstrate the importance of exploring different representations to find the appropriate operating point. We demonstrate the benefit of exploring a range of representations of network traffic and present Traffic Refinery, a proof-of-concept implementation that both monitors network traffic at 10 Gbps and transforms traffic in real time to produce a variety of feature representations for machine learning. Traffic Refinery both highlights this design space and makes it possible to explore different representations for learning, balancing systems costs related to feature extraction and model training against model accuracy.

cs.NI↗

arXiv2021

On the Deployability of Augmented Reality Using Embedded Edge Devices

Edge Computing exploits computational capabilities deployed at the very edge of the network to support applications with low latency requirements. Such capabilities can reside in small embedded devices that integrate dedicated hardware -- e.g., a GPU -- in a low cost package. But these devices have limited computing capabilities compared to standard server grade equipment. When deploying an Edge Computing based application, understanding whether the available hardware can meet target requirements is key in meeting the expected performance. In this paper, we study the feasibility of deploying Augmented Reality applications using Embedded Edge Devices (EEDs). We compare such deployment approach to one exploiting a standard dedicated server grade machine. Starting from an empirical evaluation of the capabilities of these devices, we propose a simple theoretical model to compare the performance of the two approaches. We then validate such model with NS-3 simulations and study their feasibility. Our results show that there is no one-fits-all solution. If we need to deploy high responsiveness applications, we need a centralized server grade architecture and we can in any case only support very few users. The centralized architecture fails to serve a larger number of users, even when low to mid responsiveness is required. In this case, we need to resort instead to a distributed deployment based on EEDs.

cs.DC↗

arXiv2020

Characterizing Service Provider Response to the COVID-19 Pandemic in the United States

The COVID-19 pandemic has resulted in dramatic changes to the daily habits of billions of people. Users increasingly have to rely on home broadband Internet access for work, education, and other activities. These changes have resulted in corresponding changes to Internet traffic patterns. This paper aims to characterize the effects of these changes with respect to Internet service providers in the United States. We study three questions: (1)How did traffic demands change in the United States as a result of the COVID-19 pandemic?; (2)What effects have these changes had on Internet performance?; (3)How did service providers respond to these changes? We study these questions using data from a diverse collection of sources. Our analysis of interconnection data for two large ISPs in the United States shows a 30-60% increase in peak traffic rates in the first quarter of 2020. In particular, we observe traffic downstream peak volumes for a major ISP increase of 13-20% while upstream peaks increased by more than 30%. Further, we observe significant variation in performance across ISPs in conjunction with the traffic volume shifts, with evident latency increases after stay-at-home orders were issued, followed by a stabilization of traffic after April. Finally, we observe that in response to changes in usage, ISPs have aggressively augmented capacity at interconnects, at more than twice the rate of normal capacity augmentation. Similarly, video conferencing applications have increased their network footprint, more than doubling their advertised IP address space.

cs.NI↗

arXiv2019

Inferring Streaming Video Quality from Encrypted Traffic: Practical Models and Deployment Experience

Inferring the quality of streaming video applications is important for Internet service providers, but the fact that most video streams are encrypted makes it difficult to do so. We develop models that infer quality metrics (\ie, startup delay and resolution) for encrypted streaming video services. Our paper builds on previous work, but extends it in several ways. First, the model works in deployment settings where the video sessions and segments must be identified from a mix of traffic and the time precision of the collected traffic statistics is more coarse (\eg, due to aggregation). Second, we develop a single composite model that works for a range of different services (i.e., Netflix, YouTube, Amazon, and Twitch), as opposed to just a single service. Third, unlike many previous models, the model performs predictions at finer granularity (\eg, the precise startup delay instead of just detecting short versus long delays) allowing to draw better conclusions on the ongoing streaming quality. Fourth, we demonstrate the model is practical through a 16-month deployment in 66 homes and provide new insights about the relationships between Internet "speed" and the quality of the corresponding video streams, for a variety of services; we find that higher speeds provide only minimal improvements to startup delay and resolution.

cs.NI↗