The rapid growth of AI models is placing unprecedented pressure on switch ASIC memory allocated for telemetry and flow monitoring. In this poster, we argue that the predictability of AI/ML traffic patterns can be exploited by perfect hashing techniques for accurate flow- and packet-tracking with minimal memory overhead.

Diffusion models for Text-to-Image (T2I) conditional generation have recently achieved tremendous success. Yet, aligning these models with user’s intentions still involves a laborious trial-and-error process, and this challenging alignment problem has attracted considerable attention from the research community. In this work, instead of relying on fine-grained linguistic analyses of prompts, human annotation, or auxiliary vision-language models, we use Mutual Information (MI) to guide model alignment. In brief, our method uses self-supervised fine-tuning and relies on a point-wise MI estimation between prompts and images to create a synthetic fine-tuning set for improving model alignment. Our analysis indicates that our method is superior to the state-of-the-art, yet it only requires the pre-trained denoising network of the T2I model itself to estimate MI, and a simple fine-tuning strategy that improves alignment while maintaining image quality. Code available at https://github.com/Chao0511/mitune.

Learning the right representations from complex input data is the key ability of successful machine learning (ML) models. The latter are often tailored to a specific data modality. For example, recurrent neural networks (RNNs) were designed having sequential data in mind, while convolutional neural networks (CNNs) were designed to exploit spatial correlation in images. Unlike computer vision (CV) and natural language processing (NLP), each of which targets a single well-defined modality, network ML problems often have a mixture of data modalities as input. Yet, instead of exploiting such abundance, practitioners tend to rely on sub-features thereof, reducing the problem to single modality for the sake of simplicity. In this paper, we advocate for exploiting all the modalities naturally present in network data. As a first step, we observe that network data systematically exhibits a mixture of quantities (e.g., measurements), and entities (e.g., IP addresses, names, etc.). Whereas the former are generally well exploited, the latter are often underused or poorly represented (e.g., with one-hot encoding). We propose to systematically leverage language models to learn entity representations, whenever significant sequences of such entities are historically observed. Through two diverse use-cases, we show that such entity encoding can benefit and naturally augment classic quantity-based features.

While Deep Learning (DL) technologies are a promising tool to solve networking problems that map to classification tasks, their computational complexity is still too high with respect to real-time traffic measurements requirements. To reduce the DL inference cost, we propose a novel caching paradigm, that we named approximate-key caching, which returns approximate results for lookups of selected input based on cached DL inference results. While approximate cache hits alleviate DL inference workload and increase the system throughput, they however introduce an approximation error. As such, we couple approximate-key caching with an error-correction principled algorithm, that we named auto-refresh. We analytically model our caching system performance for classic LRU and ideal caches, we perform a trace-driven evaluation of the expected performance, and we compare the benefits of our proposed approach with the state-of-the-art similarity caching – this testifies the practical interest of our proposal.

Software switches are increasingly used in network function virtualization (NFV) to route traffic between virtualized network functions (VNFs) and physical network interface cards (NICs). Understanding of alternative switch designs remains deficient, however, in the absence of a comprehensive, comparative performance analysis. In this paper, we propose a methodology intended to be fair and use it to compare the performance of seven state-of-the-art software switches. We first explore their respective design spaces and then compare their performance under four representative test scenarios. Each scenario corresponds to a specific case of routing NFV traffic between NICs and/or VNFs. Our experimental results show that no single software switch prevails in all scenarios. It is therefore important to choose the one that is best adapted to a given use-case. The presented results and analysis bring a better understanding of design tradeoffs and identify potential bottlenecks that limit the performance of software switches.

Interest in software networking has grown significantly since the introduction of Network Function Virtualization (NFV). Software switches are used in NFV to steer traffic between different virtualized network functions and physical Network Interface Cards (NICs). It is becoming more and more important to objectively evaluate and compare the performance of the multiple alternative implementations that have recently been proposed. A comprehensive performance analysis is still missing for two main reasons: (i) the amount of time required to configure and compare all such tools is enormous; (ii) it is very difficult to define a proper methodology to compare different solutions in a fair manner. In this paper we propose a methodology based on four simple yet representative test scenarios used to evaluate the performance of software switches. We apply this methodology to measure throughput and latency metrics for 6 state-of-the-art software switches namely, OVS-DPDK, snabb, BESS, FastClick, VPP and netmap VALE. Our work constitutes a first step to building a better understanding of design tradeoffs and identifying performance bottlenecks.

In the last few years, several software-based solutions have been proved to be very efficient for high-speed packet processing, traffic generation, and monitoring, and can be considered valid alternatives to expensive and non-flexible hardware-based solutions. In this paper, we first benchmark heterogeneous design choices for software-based packet monitoring systems in terms of achievable performance and required resources (i.e., the number of CPU cores). Building on this extensive analysis we design FloWatcher-DPDK, a DPDK-based high-speed software traffic monitor we provide to the community as an open source project. In a nutshell, FloWatcher-DPDK provides tunable fine-grained statistics at packet and flow levels. Experimental results demonstrate that FloWatcher-DPDK sustains per-flow statistics with 5-nines precision at high-speed (e.g., 14.88 Mpps) using a limited amount of resources. Finally, we showcase the usage of FloWatcher-DPDK by configuring it to analyze the performance of two open source prototypes for stateful flow-level end-host and in-network packet processing.

The upcoming 5G architecture is expected to heavily rely on network functions implemented by software deployed on commodity hardware architectures. Multiple standardization efforts are underway to specify interfaces between virtualized and real infrastructure, and procedures for interoperability among functions. However, the practical feasibility of function implementation in such abstract and disembodied conditions is scarcely covered in the latest literature. In this paper, we argue for a Network Function Virtualization (NFV) framework that provides 5G network functions built around a modular software router model, rather than following the traditional VM-container approaches. We illustrate its advantages in enabling support for efficient processing on heterogeneous hardware and in ensuring consistency of flow/session semantics across distributed 5G data planes. Finally, we report on the state of Programmable Data Plane, our architecture to implement 5G network functions as modular pipelines orchestrated across multiple devices.

The rapid traffic growth fueled by mobile devices spread and high speed network access calls for substantial innovation at network layer. The content-centric nature of Internet usage highlights the limitations of the host-centric model in coping with dynamic content-to-location binding, mobility, multicast, multi-homing, etc. If transmission capacity speedups in the backhaul may hide inefficiencies in the short term, the hostcentric communication model needs to be revisited to sustain future mobile demand. In this paper, we first identify and quantify the opportunities for backhaul evolution by analyzing a large set of traffic measurements collected between mobile core and backhaul of Orange France. The analysis reveals that 50% of HTTP requests are cacheable and traffic can be reduced from 60% to 95% during the peak hour by using 350GBs to 1TB of memory overall. Motivated by such significant opportunities for latency reduction and network cost savings, we present a solution based on Information-Centric Networking (ICN). First results of a large scale experimentation with 100 Linux servers and customized software, in a realistic network setting, provide a glimpse into ICN gains even under naive caching: a factor three reduction in delivery time and almost 40% bandwidth savings, when compared to existing alternatives.

The evolution of the Internet into a distributed Information access system calls for a paradigm shift to enable an evolvable future network architecture. Information-Centric Networking (ICN) proposals rethink the communication model around named data, in contrast with the host-centric transport view of TCP/IP. Information retrieval is natively pull-based, driven by user requests, point-to-multipoint and intrinsically coupled with in-network caching. In this paper, we tackle the problem of joint multipath congestion control and request forwarding in ICN for the first time. We formulate it as a global optimization problem with the twofold objective of maximizing user throughput and minimizing overall network cost. We solve it via decomposition and derive a family of optimal congestion control strategies at the receiver and of distributed algorithms for dynamic request forwarding at network nodes. An experimental evaluation of our proposal is carried out in different network scenarios to assess the performance of our design and to highlight the benefits of an ICN approach.

Caching is a key component for Content Distribution Networks and new Information-Centric Network architectures. In this paper, we address performance issues of caching networks running the RND replacement policy. We first prove that when the popularity distribution follows a general power-law with decay exponent α > 1, the miss probability is asymptotic to O( C1-α) for large cache size C. We further evaluate network of caches under RND policy for homogeneous tree networks and extend the analysis to tandem cache networks where caches employ either LRU or RND policies.

In this paper we define a simple experimental setup to analyze the behavior of commercial P2P-TV applications under adverse network conditions. Our goal is to reveal the ability of different P2P-TV applications to adapt to dynamically changing conditions, such as delay, loss and available capacity, e.g., checking whether such systems implement some form of congestion control. We apply our methodology to four popular commercial P2P-TV applications: PPLive, SOPCast, TVants and TVUPlayer. Our results show that all the considered applications are in general capable to cope with packet losses and to react to congestion arising in the network core. Indeed, all applications keep trying to download data by avoiding bad paths and carefully selecting good peers. However, when the bottleneck affects all peers, e.g., it is at the access link, their behavior results rather aggressive, and potentially harmful for both other applications and the network.