arXiv subjects

Jose Yallouz

Publications and source records attributed to Jose Yallouz.

6 recordsLinked to original sources

arXiv2026

Incast-Free MoE Rate-Based Scheduling

Mixture of Experts (MoE) architectures have become key to large language models; however, their typical round-robin (RR) scheduling introduces significant bottlenecks. In this paper, we demonstrate that RR causes a previously-undiscovered exponential incast phenomenon with MoE traffic. We propose an alternative proactive fair scheduling framework tailored for MoE workloads, which effectively prevents fabric oversubscription. We also outline how it can be implemented in NICs. Finally, through extensive simulations with real and synthetic workloads, we demonstrate that this framework consistently eliminates incast, maintains a near-100% link utilization, and reduces Collective Completion Time (CCT).

cs.NI↗

arXiv2026

Learning-Augmented Scalable Linear Assignment Problem Optimization via Neural Dual Warm-Starts

The Linear Assignment Problem is a fundamental combinatorial optimization task where classical exact solvers ensure optimality but suffer from an $\mathcal{O}(N^{3})$ bottleneck, while recent neural approximations struggle with scalability and exactness. We propose a learning-augmented framework that accelerates exact solvers by predicting dual variables to warm-start the search, backed by a fallback mechanism to preserve worst-case guarantees. Central to our approach is RowDualNet, a lightweight, row-independent architecture that avoids the $\mathcal{O}(N^{2})$ memory bottleneck of graph models, enabling scalable neural warm-starting up to $N=16{,}384$. Feasibility is guaranteed by construction via the Min-Trick mechanism, completely eliminating the need for costly iterative projections. Empirically, our method drastically reduces the search effort of the Jonker-Volgenant (LAPJV) algorithm, yielding robust zero-shot generalization with strict optimality and end-to-end speedups of over 2x on complex synthetic data, 1.25x on real-world tracking, and 1.5x on transportation networks.

cs.LG↗

arXiv2026

Scaling Routers with In-Package Optics and High-Bandwidth Memories

This paper aims to apply two major scaling transformations from the computing packaging industry to internet routers: the heterogeneous integration of high-bandwidth memories (HBMs) and chiplets, as well as in-package optics. We propose a novel internet router architecture that employs these technologies to achieve a petabit/sec router within a single integrated package. At the top-level, we introduce a novel split-parallel switch architecture that spatially divides (without processing) the incoming fibers and distributes them across smaller independent switches without intermediate OEO conversions or fine-tuned per-packet load-balancing. This passive spatial division enables scaling at the cost of a coarser traffic load balancing. Yet, through extensive evaluations of backbone network traffic, we demonstrate that differences with fine-tuned approaches are small. In addition, we propose a novel HBM-based shared-memory architecture for the implementation of the smaller independent switches, and we introduce a novel parallel frame interleaving algorithm that packs traffic into frames so that HBM banks are accessed at peak HBM data rates in a cyclical interleaving manner. We further discuss why these new technologies represent a paradigm shift in the design of future internet routers. Finally, we emphasize that power consumption may constitute the primary bottleneck to scaling.

cs.NI↗

arXiv2022

Using Internet Measurements to Map the 2022 Ukrainian Refugee Crisis

The conflict in Ukraine, starting in February 2022, began the largest refugee crisis in decades, with millions of Ukrainian refugees crossing the border to neighboring countries and millions of others forced to move within the country. In this paper we present an insight into how Internet measurements can be used to analyze the refugee crisis. Based on preliminary data from the first two months of the war we analyze how measurement data indicates the trends in the flow of refugees from Ukraine to its neighboring countries, and onward to other countries. We believe that these insights can greatly contribute to the ongoing international effort to map the flow of refugees in order to aid and protect them.

cs.NI↗

arXiv2022

Internet Performance in the 2022 Conflict in Ukraine: An Asymmetric Analysis

On 24 February 2022 Russia invaded Ukraine, starting one of the largest military conflicts in Europe in recent years. In this paper we present preliminary findings about the impact of the conflict on the Internet performance in Ukraine and in Russia, introducing an ironically asymmetric picture: the Internet performance in Ukraine has significantly degraded, while the performance in Russia has improved.

cs.NI↗

arXiv2016

Tunable QoS-Aware Network Survivability

Coping with network failures has been recognized as an issue of major importance in terms of social security, stability and prosperity. It has become clear that current networking standards fall short of coping with the complex challenge of surviving failures. The need to address this challenge has become a focal point of networking research. In particular, the concept of \textbf{\emph{tunable survivability}} offers major performance improvements over traditional approaches. Indeed, while the traditional approach aims at providing full (100\%) protection against network failures through disjoint paths, it was realized that this requirement is too restrictive in practice. Tunable survivability provides a quantitative measure for specifying the desired level (0\%-100\%) of survivability and offers flexibility in the choice of the routing paths. Previous work focused on the simpler class of "bottleneck" criteria, such as bandwidth. In this study, we focus on the important and much more complex class of \emph{additive} criteria, such as delay and cost. First, we establish some (in part, counter-intuitive) properties of the optimal solution. Then, we establish efficient algorithmic schemes for optimizing the level of survivability under additive end-to-end QoS bounds. Subsequently, through extensive simulations, we show that, at the price of \emph{negligible} reduction in the level of survivability, a major improvement (up to a factor of $2$) is obtained in terms of end-to-end QoS performance. Finally, we exploit the above findings in the context of a network design problem, in which, for a given investment budget, we aim to improve the survivability of the network links.

cs.NI↗