arXiv subjects

Haryadi S. Gunawi

Publications and source records attributed to Haryadi S. Gunawi.

6 recordsLinked to original sources

arXiv2026

A Year in LLM Serving: Workload Evolution, Caching and Load-Balancing

Large Language Model (LLM) serving has become a critical cloud workload, and realistic traces are essential for motivating and benchmarking serving systems. However, existing LLM serving workload studies remain limited in scale and scope. They often observe short time periods and provide limited visibility into how users interact with models in production. As a result, they do not fully capture how LLM serving workloads evolve over time or how user-model interactions shape production traffic. In this work, we further the understanding of real-world LLM serving workloads through both a global characterization and a longitudinal study of a one-year production trace from CompanyX. Unlike prior studies, our trace captures full production behavior across many models and users, including both popular and long-tail models. We analyze the workload from aggregate, temporal, model-level, and user-level perspectives, revealing workload evolution and user-model structure that are typically hidden behind aggregate views. To support future research, we publicly release the full one-year trace, enabling downstream studies of production behavior without relying on sampled or synthetically generated workloads. The trace is available at https://github.com/HarvardMadSys/chutes_workload.

cs.AI↗

arXiv2026

Understanding and Detecting Scalability Faults in Large-Scale Distributed Systems

Scalable distributed systems form the backbone of modern computing infrastructure. However, as scale grows, system complexity may lead to scalability faults. Scalability faults are challenging to uncover and diagnose, as they are often latent and only manifest at large-scale deployment. In this paper, we present the first comprehensive study on scalability faults and propose an approach for their detection. First, we systematically investigate 444 scalability issue reports from 10 large-scale distributed systems to understand the common anti-patterns and root causes of scalability faults. We found that the majority of these faults are caused by the synergy between dimensional code fragments and anti-patterns associated with them. Second, based on our findings, we design and implement ScaleLens, a novel approach to detect scalability faults. ScaleLens combines dynamic and static analyses to pinpoint dimensional code fragments and match them with anti-patterns. Our evaluation shows that ScaleLens detects 4.2x more dimensional code fragments associated with known scalability faults compared to the baseline. On the latest stable versions of Cassandra, HDFS, and Ignite, ScaleLens detects 334 dimensional code fragments with confirmed problematic behavior.

cs.SE↗

arXiv2026

StorRep: Storage Research Experiment Patterns on Chameleon Cloud and Trovi

Storage experiments are vital to advancing storage research, but creating extensible and reproducible storage artifacts can be a challenging task. Our research has shown that only 1% of SSD simulator-based experiences are packaged and 0.5% of them can be easily reproduced. The lack of such artifacts without proper reproducibility can significantly impede the advancement of storage research. The biggest challenges in these types of experiments are ensuring that we have the correct environment to conduct them and creating extensible experiments that can be built upon in future research. To address this issue, we introduce StorRep, a thorough study that provides six extensible and reproducible storage experiment artifacts that serve as the foundation for further storage research, utilizing the Chameleon infrastructure. Our study offers experiment patterns and guidelines that can help researchers create transparent and dependable storage experiments. We have successfully integrated our methods in several experiments in multiple community and educational events over several years and produced publicly accessible artifacts that can be extended and fully reproduced without any restrictions.

cs.DC↗

arXiv2025

Alchemist: Towards the Design of Efficient Online Continual Learning System

Continual learning has become a promising solution to refine large language models incrementally by leveraging user feedback. In particular, online continual learning - iteratively training the model with small batches of user feedback - has demonstrated notable performance improvements. However, the existing practice of separating training and serving processes forces the online trainer to recompute the intermediate results already done during serving. Such redundant computations can account for 30%-42% of total training time. In this paper, we propose Alchemist, to the best of our knowledge, the first online continual learning system that efficiently reuses serving activations to increase training throughput. Alchemist introduces two key techniques: (1) recording and storing activations and KV cache only during the prefill phase to minimize latency and memory overhead; and (2) smart activation offloading and hedging. Evaluations with inputs of varied token length sampled from ShareGPT dataset show that compared with a separate training cluster, Alchemist significantly increases training throughput by up to 1.72x, reduces up to 47% memory usage during training, and supports up to 2x more training tokens - all while maintaining negligible impact on serving latency.

cs.LG↗

arXiv2023

Towards Continually Learning Application Performance Models

Machine learning-based performance models are increasingly being used to build critical job scheduling and application optimization decisions. Traditionally, these models assume that data distribution does not change as more samples are collected over time. However, owing to the complexity and heterogeneity of production HPC systems, they are susceptible to hardware degradation, replacement, and/or software patches, which can lead to drift in the data distribution that can adversely affect the performance models. To this end, we develop continually learning performance models that account for the distribution drift, alleviate catastrophic forgetting, and improve generalizability. Our best model was able to retain accuracy, regardless of having to learn the new distribution of data inflicted by system changes, while demonstrating a 2x improvement in the prediction accuracy of the whole data sequence in comparison to the naive approach.

cs.LG↗

arXiv2013

Impact of Limpware on HDFS: A Probabilistic Estimation

With the advent of cloud computing, thousands of machines are connected and managed collectively. This era is confronted with a new challenge: performance variability, primarily caused by large-scale management issues such as hardware failures, software bugs, and configuration mistakes. In our previous work we highlighted one overlooked cause: limpware - hardware whose performance degrades significantly compared to its specification. We showed that limpware can cause severe impact in current scale-out systems. In this report, we quantify how often these scenarios happen in Hadoop Distributed File System.

cs.OS↗