arXiv subjects

Hanoch Levy

Publications and source records attributed to Hanoch Levy.

6 recordsLinked to original sources

arXiv2026

Evaluating Large Language Models on Computer Science University Exams in Data Structures

We present a comprehensive evaluation of Large Language Models (LLMs) on Computer Science (CS) Data Structure examination questions. Our work introduces a new benchmark dataset comprising exam questions from Tel Aviv University (TAU), curated to assess LLMs' abilities in handling closed and multiple-choice questions. We evaluated the performance of OpenAI's GPT 4o and Anthropic's Claude 3.5, popular LLMs, alongside two smaller LLMs, Mathstral 7B and LLaMA 3 8B, across the TAU exams benchmark. Our findings provide insight into the current capabilities of LLMs in CS education.

cs.CL↗

arXiv2025

RetryGuard: Preventing Self-Inflicted and Attack-Driven Retry Storms in Cloud Applications

Modern cloud applications are built from independent microservices, offering scalability and usage-based billing. However, their reliance on independently-operating auto-scalers introduces coordination challenges. Default retry patterns can trigger "retry storms" during service miscoordination or adversarial overload, amplifying load, latency, and resource billing. These storms can cause either self-inflicted Denial-of-Wallet (DoW) or amplify the impact of DDoS attacks. To overcome these problems, we introduce RetryGuard, a distributed framework for productive control of retry patterns across interdependent microservices. By managing retry policy on a per-service basis and making parallel decisions, RetryGuard prevents retry storms, curbs resource contention, and mitigates escalating operational costs. RetryGuard makes its decisions based on an analytical model that captures the relationships among retries, throughput (rejections), delays, and costs. Simulations show that it outperforms established mechanisms, including exponential backoff, jitter, and retry budgets. Experimental results show that RetryGuard significantly reduces resource usage and costs compared to existing AWS policies, achieving more than 90% improvement in latency, and 98% reduction in storm size. We further demonstrate its effectiveness in a multi-layer Kubernetes deployment with the Istio service-mesh, where RetryGuard reduces the peak number of replicas by $3 \times$ and cumulative memory usage by 55%.

cs.NI↗

arXiv2024

Resilience of Networks to Spreading Computer Viruses: Optimal Anti-Virus Deployment (Extended Version)

Deployment of anti-virus software is a common strategy for preventing and controlling the propagation of computer viruses and worms over a computer network. As the deployment of such programs is often limited due to monetary or operational costs, devising optimal strategies for their allocation and deployment can be of high value to the operation, performance, and resilience of the target networks. We study the effects of anti-virus deployment (i.e., "vaccination") strategies on the ability of a network to block the spread of a virus. Such ability is obtained when the network reaches "herd immunity", achieved when a large fraction of the network entities is immune to the infection, which provides protection even for entities which are not immune. We use a model that explicitly accounts for the inherent heterogeneity of network nodes activity and derive optimal strategies for anti-virus deployment. Numerical evaluations demonstrate that the system performance is very sensitive to the chosen strategy, and thus strategies which disregard the heterogeneous spread nature may perform significantly worse relatively to those derived in this work.

cs.NI↗

arXiv2022

On The Convexity Of The Effective Reproduction Number

In this study we analyze the evolution of the effective reproduction number, $R$, through a SIR spreading process in heterogeneous networks; Characterizing its decay process allows to analytically study the effects of countermeasures on the progress of the virus under heterogeneity, and to optimize their policies. A striking result of recent studies has shown that heterogeneity across nodes/individuals (or, super-spreading) may have a drastic effect on the spreading process progression, which may cause a non-linear decrease of $R$ in the number of infected individuals. We account for heterogeneity and analyze the stochastic progression of the spreading process. We show that the decrease of $R$ is, in fact, convex in the number of infected individuals, where this convexity stems from heterogeneity. The analysis is based on establishing stochastic monotonic relations between the susceptible populations in varying times of the spread. We demonstrate that the convex behavior of the effective reproduction number affects the performance of countermeasures used to fight a spread of a virus. The results are applicable to the control of virus and malware spreading in computer networks as well. We examine numerically the sensitivity of the Herd Immunity Threshold (HIT) to the heterogeneity level and to the chosen policy.

cs.SI↗

arXiv2021

Super-Spreaders Out, Super-Spreading In: The Effects of Infectiousness Heterogeneity and Lockdowns on Herd Immunity

Recently, [8] has proposed that heterogeneity of infectiousness (and susceptibility) across individuals in infectious diseases, plays a major role in affecting the Herd Immunity Threshold (HIT). Such heterogeneity has been observed in COVID-19 and is recognized as overdispersion (or "super-spreading"). The model of [8] suggests that super-spreaders contribute significantly to the effective reproduction factor, R, and that they are likely to get infected and immune early in the process. Consequently, under R_0 = 3 (attributed to COVID-19), the Herd Immunity Threshold (HIT) is as low as 5%, in contrast to 67% according to the traditional models [1, 2, 4, 10]. This work follows up on [8] and proposes that heterogeneity of infectiousness (susceptibility) has two "faces" whose mix affects dramatically the HIT: (1) Personal-Trait-, and (2) Event-Based- Infectiousness (Susceptibility). The former is a personal trait of specific individuals (super-spreaders) and is nullified once those individuals are immune (as in [8]). The latter is event-based (e.g cultural super-spreading events) and remains effective throughout the process, even after the super-spreaders immune. We extend [8]'s model to account for these two factors, analyze it and conclude that the HIT is very sensitive to the mix between (1) and (2), and under R_0 = 3 it can vary between 5% and 67%. Preliminary data from COVID-19 suggests that herd immunity is not reached at 5%. We address operational aspects and analyze the effects of lockdown strategies on the spread of a disease. We find that herd immunity (and HIT) is very sensitive to the lockdown type. While some lockdowns affect positively the disease blocking and increase herd immunity, others have adverse effects and reduce the herd immunity.

q-bio.PE↗

arXiv2009

On Leveraging Partial Paths in Partially-Connected Networks

Mobile wireless network research focuses on scenarios at the extremes of the network connectivity continuum where the probability of all nodes being connected is either close to unity, assuming connected paths between all nodes (mobile ad hoc networks), or it is close to zero, assuming no multi-hop paths exist at all (delay-tolerant networks). In this paper, we argue that a sizable fraction of networks lies between these extremes and is characterized by the existence of partial paths, i.e. multi-hop path segments that allow forwarding data closer to the destination even when no end-to-end path is available. A fundamental issue in such networks is dealing with disruptions of end-to-end paths. Under a stochastic model, we compare the performance of the established end-to-end retransmission (ignoring partial paths), against a forwarding mechanism that leverages partial paths to forward data closer to the destination even during disruption periods. Perhaps surprisingly, the alternative mechanism is not necessarily superior. However, under a stochastic monotonicity condition between current v.s. future path length, which we demonstrate to hold in typical network models, we manage to prove superiority of the alternative mechanism in stochastic dominance terms. We believe that this study could serve as a foundation to design more efficient data transfer protocols for partially-connected networks, which could potentially help reducing the gap between applications that can be supported over disconnected networks and those requiring full connectivity.

cs.NI↗