arXiv subjects

Behnaz Arzani

Publications and source records attributed to Behnaz Arzani.

16 recordsLinked to original sources

arXiv2026

A Performance Analyzer for a Public Cloud's ML-Augmented VM Allocator

Cloud operators increasingly deploy multiple ML models in their VM allocation pipelines. In such settings, individually benign predictions can shift and compound, severely degrading performance. In a cloud provider's VM placement pipeline, CPU, memory, and lifetime prediction models jointly determine server count, live migration frequency, and network utilization; yet no existing approach can systematically stress-test how these models adversely interact. Deterministic adversarial analyzers cannot capture probabilistic ML behavior, so operators miss failures that arise only from correlated distributional shifts across models In SANJESH, we formulate a bi-level optimization that captures how the ML models behave statistically and uncovers how they adversely interact. The outer level searches over what predictions the ML models could produce under distributional uncertainty to find adversarial conditions; the inner level evaluates how the VM allocator behaves given those predictions. When we applied it to the operator's production traces, SANJESH uncovered scenarios that cause $4\times$ worse performance than the operators' evaluator detected.

cs.DC↗

arXiv2026

Harvest: Adaptive Photonic Switching Schedules for Collective Communication in Scale-up Domains

As chip-to-chip silicon photonics gain traction for their bandwidth and energy efficiency, their circuit-switched nature raises a fundamental question for collective communication: when and how should the interconnect be reconfigured to realize these benefits? Establishing direct optical paths can reduce congestion and propagation delay, but each reconfiguration incurs non-negligible overhead, making naive per-step reconfiguration impractical. We present Harvest, a systematic approach for synthesizing topology reconfiguration schedules that minimize collective completion time in photonic interconnects. Given a collective communication algorithm and its fixed communication schedule, Harvest determines how the interconnect should evolve over the course of the collective, explicitly balancing reconfiguration delay against congestion and propagation delay. We reduce the synthesis problem into a dynamic program with an underlying topology optimization subproblem and show that the approach applies to arbitrary collective communication algorithms. Furthermore, we exploit the algorithmic structure of a well-known AllReduce algorithm (Recursive Doubling) to synthesize optimal reconfiguration schedules without using any optimizers. By parameterizing the formulation using reconfiguration delay, Harvest naturally adapts to various photonic technologies. Using packet-level and flow-level evaluations, as well as hardware emulation on commercial GPUs, we show that the schedules synthesized by Harvest significantly reduce collective completion time across multiple collective algorithms compared to static interconnects and reconfigure-every-step baselines.

cs.NI↗

arXiv2025

Dynamic Rebatching for Efficient Early-Exit Inference with DREX

Early-Exit (EE) is a Large Language Model (LLM) architecture that accelerates inference by allowing easier tokens to be generated using only a subset of the model's layers. However, traditional batching frameworks are ill-suited for EE LLMs, as not all requests in a batch may be ready to exit at the same time. Existing solutions either force a uniform decision on the batch, which overlooks EE opportunities, or degrade output quality by forcing premature exits. We propose Dynamic Rebatching, a solution where we dynamically reorganize the batch at each early-exit point. Requests that meet the exit criteria are immediately processed, while those that continue are held in a buffer, re-grouped into a new batch, and forwarded to deeper layers. We introduce DREX, an early-exit inference system that implements Dynamic Rebatching with two key optimizations: 1) a copy-free rebatching buffer that avoids physical data movement, and 2) an EE and SLA-aware scheduler that analytically predicts whether a given rebatching operation will be profitable. DREX also efficiently handles the missing KV cache from skipped layers using memory-efficient state-copying. Our evaluation shows that DREX improves throughput by 2-12% compared to baseline approaches while maintaining output quality. Crucially, DREX completely eliminates involuntary exits, providing a key guarantee for preserving the output quality intended by the EE model.

cs.DC↗

arXiv2025

Robust Heuristic Algorithm Design with LLMs

We posit that we can generate more robust and performant heuristics if we augment approaches using LLMs for heuristic design with tools that explain why heuristics underperform and suggestions about how to fix them. We find even simple ideas that (1) expose the LLM to instances where the heuristic underperforms; (2) explain why they occur; and (3) specialize design to regions in the input space, can produce more robust algorithms compared to existing techniques~ -- ~the heuristics we produce have a $\sim28\times$ better worst-case performance compared to FunSearch, improve average performance, and maintain the runtime.

cs.AI↗

arXiv2025

Enhancing Network Failure Mitigation with Performance-Aware Ranking

Cloud providers install mitigations to reduce the impact of network failures within their datacenters. Existing network mitigation systems rely on simple local criteria or global proxy metrics to determine the best action. In this paper, we show that we can support a broader range of actions and select more effective mitigations by directly optimizing end-to-end flow-level metrics and analyzing actions holistically. To achieve this, we develop novel techniques to quickly estimate the impact of different mitigations and rank them with high fidelity. Our results on incidents from a large cloud provider show orders of magnitude improvements in flow completion time and throughput. We also show our approach scales to large datacenters.

cs.NI↗

arXiv2024

Towards Safer Heuristics With XPlain

Many problems that cloud operators solve are computationally expensive, and operators often use heuristic algorithms (that are faster and scale better than optimal) to solve them more efficiently. Heuristic analyzers enable operators to find when and by how much their heuristics underperform. However, these tools do not provide enough detail for operators to mitigate the heuristic's impact in practice: they only discover a single input instance that causes the heuristic to underperform (and not the full set), and they do not explain why. We propose XPlain, a tool that extends these analyzers and helps operators understand when and why their heuristics underperform. We present promising initial results that show such an extension is viable.

cs.AI↗

arXiv2023

Finding Adversarial Inputs for Heuristics using Multi-level Optimization

Production systems use heuristics because they are faster or scale better than their optimal counterparts. Yet, practitioners are often unaware of the performance gap between a heuristic and the optimum or between two heuristics in realistic scenarios. We present MetaOpt, a system that helps analyze heuristics. Users specify the heuristic and the optimal (or another heuristic) as input, and MetaOpt automatically encodes these efficiently for a solver to find performance gaps and their corresponding adversarial inputs. Its suite of built-in optimizations helps it scale its analysis to practical problem sizes. To show it is versatile, we used MetaOpt to analyze heuristics from three domains (traffic engineering, vector bin packing, and packet scheduling). We found a production traffic engineering heuristic can require 30% more capacity than the optimal to satisfy realistic demands. Based on the patterns in the adversarial inputs MetaOpt produced, we modified the heuristic to reduce its performance gap by 12.5$\times$. We examined adversarial inputs to a vector bin packing heuristic and proved a new lower bound on its performance.

cs.NI↗

arXiv2023

Solving Max-Min Fair Resource Allocations Quickly on Large Graphs

We consider the max-min fair resource allocation problem. The best-known solutions use either a sequence of optimizations or waterfilling, which only applies to a narrow set of cases. These solutions have become a practical bottleneck in WAN traffic engineering and cluster scheduling, especially at larger problem sizes. We improve both approaches: (1) we show how to convert the optimization sequence into a single fast optimization, and (2) we generalize waterfilling to the multi-path case. We empirically show our new algorithms Pareto-dominate prior techniques: they produce faster, fairer, and more efficient allocations. Some of our allocators also have theoretical guarantees: they trade off a bounded amount of unfairness for faster allocation. We have deployed our allocators in Azure's WAN traffic engineering pipeline, where we preserve solution quality and achieve a roughly $3\times$ speedup.

cs.NI↗

arXiv2023

Rethinking Machine Learning Collective Communication as a Multi-Commodity Flow Problem

We show communication schedulers' recent work proposed for ML collectives does not scale to the increasing problem sizes that arise from training larger models. These works also often produce suboptimal schedules. We make a connection with similar problems in traffic engineering and propose a new method, TECCL, that finds better quality schedules (e.g., finishes collectives faster and/or while sending fewer bytes) and does so more quickly on larger topologies. We present results on many different GPU topologies that show substantial improvement over the state-of-the-art.

cs.NI↗

arXiv2022

FedSpace: An Efficient Federated Learning Framework at Satellites and Ground Stations

Large-scale deployments of low Earth orbit (LEO) satellites collect massive amount of Earth imageries and sensor data, which can empower machine learning (ML) to address global challenges such as real-time disaster navigation and mitigation. However, it is often infeasible to download all the high-resolution images and train these ML models on the ground because of limited downlink bandwidth, sparse connectivity, and regularization constraints on the imagery resolution. To address these challenges, we leverage Federated Learning (FL), where ground stations and satellites collaboratively train a global ML model without sharing the captured images on the satellites. We show fundamental challenges in applying existing FL algorithms among satellites and ground stations, and we formulate an optimization problem which captures a unique trade-off between staleness and idleness. We propose a novel FL framework, named FedSpace, which dynamically schedules model aggregation based on the deterministic and time-varying connectivity according to satellite orbits. Extensive numerical evaluations based on real-world satellite images and satellite networks show that FedSpace reduces the training time by 1.7 days (38.6%) over the state-of-the-art FL algorithms.

cs.LG↗

arXiv2021

Towards a Cost vs. Quality Sweet Spot for Monitoring Networks

Continuously monitoring a wide variety of performance and fault metrics has become a crucial part of operating large-scale datacenter networks. In this work, we ask whether we can reduce the costs to monitor -- in terms of collection, storage and analysis -- by judiciously controlling how much and which measurements we collect. By positing that we can treat almost all measured signals as sampled time-series, we show that we can use signal processing techniques such as the Nyquist-Shannon theorem to avoid wasteful data collection. We show that large savings appear possible by analyzing tens of popular measurements from a production datacenter network. We also discuss the technical challenges that must be solved when applying these techniques in practice.

cs.NI↗

arXiv2021

Interpret-able feedback for AutoML systems

Automated machine learning (AutoML) systems aim to enable training machine learning (ML) models for non-ML experts. A shortcoming of these systems is that when they fail to produce a model with high accuracy, the user has no path to improve the model other than hiring a data scientist or learning ML -- this defeats the purpose of AutoML and limits its adoption. We introduce an interpretable data feedback solution for AutoML. Our solution suggests new data points for the user to label (without requiring a pool of unlabeled data) to improve the model's accuracy. Our solution analyzes how features influence the prediction among all ML models in an AutoML ensemble, and we suggest more data samples from feature ranges that have high variance in such analysis. Our evaluation shows that our solution can improve the accuracy of AutoML by 7-8% and significantly outperforms popular active learning solutions in data efficiency, all the while providing the added benefit of being interpretable.

cs.LG↗

arXiv2020

Towards A Domain-Customized Automated Machine Learning Framework For Networks and Systems

Clouds gather a vast volume of telemetry from their networked systems which contain valuable information that can help solve many of the problems that continue to plague them. However, it is hard to extract useful information from such raw data. Machine Learning (ML) models are useful tools that enable operators to either leverage this data to solve such problems or develop intuition about whether/how they can be solved. Building practical ML models is time-consuming and requires experts in both ML and networked systems to tailor the model to the system/network (a.k.a "domain-customize" it). The number of applications we deploy exacerbates the problem. The speed with which our systems evolve and with which new monitoring systems are deployed (deprecated) means these models often need to be adapted to keep up. Today, the lack of individuals with both sets of expertise is becoming one of the bottlenecks for adopting ML in cloud operations. This paper argues it is possible to build a domain-customized automated ML framework for networked systems that can help save valuable operator time and effort.

cs.NI↗

arXiv2018

Sunstar: A Cost-effective Multi-Server Solution for Reliable Video Delivery

In spite of much progress and many advances, cost-effective, high-quality video delivery over the internet remains elusive. To address this ongoing challenge, we propose Sunstar, a solution that leverages simultaneous downloads from multiple servers to preserve video quality. The novelty in Sunstar is not so much in its use of multiple servers but in the design of a schedule that balances improvements in video quality against increases in (peering) costs. The paper's main contributions are in elucidating the impact on the cost of various approaches towards improving video quality, including the use of multiple servers, and incorporating this understanding into the design of a scheduler capable of realizing an efficient trade-off. Our results show that Sunstar's scheduling algorithm can significantly improve performance (up to 50% in some instances) without cost impacts.

cs.NI↗

arXiv2018

007: Democratically Finding The Cause of Packet Drops

Network failures continue to plague datacenter operators as their symptoms may not have direct correlation with where or why they occur. We introduce 007, a lightweight, always-on diagnosis application that can find problematic links and also pinpoint problems for each TCP connection. 007 is completely contained within the end host. During its two month deployment in a tier-1 datacenter, it detected every problem found by previously deployed monitoring tools while also finding the sources of other problems previously undetected.

cs.NI↗

arXiv2015

FixRoute: A Unified Logic and Numerical Tool for Provably Safe Internet Traffic Engineering

The performance of networks that use the Internet Protocol is sensitive to precise configuration of many low-level parameters on each network device. These settings govern the action of dynamic routing protocols, which direct the flow of traffic; in order to ensure that these dynamic protocols all converge to produce some 'optimal' flow, each parameter must be set correctly. Multiple conflicting optimization objectives, nondeterminism, and the need to reason about different failure scenarios make the task particularly complicated. We present a fast and flexible approach for the analysis of a number of such management tasks presented in the context of BGP routing. The idea is to combine {\em logical} satisfiability criteria with traditional {\em numerical} optimization, to reach a desired traffic flow outcome subject to given constraints on the routing process. The method can then be used to probe parameter sensitivity, trade-offs in the selection of optimization goals, resilience to failure, and so forth. The theory is underpinned by a rigorous abstraction of the convergence of distributed asynchronous message-passing protocols, and is therefore generalizable to other scenarios. Our resulting hybrid engine is faster than either purely logical or purely numeric alternatives, making it potentially feasible for interactive production use.

cs.NI↗