arXiv subjects

Martin Maas

Publications and source records attributed to Martin Maas.

9 recordsLinked to original sources

arXiv2025

Instruction Set Migration at Warehouse Scale

Migrating codebases from one instruction set architecture (ISA) to another is a major engineering challenge. A recent example is the adoption of Arm (in addition to x86) across the major Cloud hyperscalers. Yet, this problem has seen limited attention by the academic community. Most work has focused on static and dynamic binary translation, and the traditional conventional wisdom has been that this is the primary challenge. In this paper, we show that this is no longer the case. Modern ISA migrations can often build on a robust open-source ecosystem, making it possible to recompile all relevant software from scratch. This introduces a new and multifaceted set of challenges, which are different from binary translation. By analyzing a large-scale migration from x86 to Arm at Google, spanning almost 40,000 code commits, we derive a taxonomy of tasks involved in ISA migration. We show how Google automated many of the steps involved, and demonstrate how AI can play a major role in automatically addressing these tasks. We identify tasks that remain challenging and highlight research challenges that warrant further attention.

cs.SE↗

arXiv2025

ECO: An LLM-Driven Efficient Code Optimizer for Warehouse Scale Computers

With the end of Moore's Law, optimizing code for performance has become paramount for meeting ever-increasing compute demands, particularly in hyperscale data centers where even small efficiency gains translate to significant resource and energy savings. Traditionally, this process requires significant programmer effort to identify optimization opportunities, modify the code to implement the optimization, and carefully deploy and measure the optimization's impact. Despite a significant amount of work on automating program edits and promising results in small-scale settings, such performance optimizations have remained elusive in large real-world production environments, due to the scale, high degree of complexity, and reliability required. This paper introduces ECO (Efficient Code Optimizer), a system that automatically refactors source code to improve performance at scale. To achieve these performance gains, ECO searches through historical commits at scale to create a dictionary of performance anti-patterns that these commits addressed. These anti-patterns are used to search for similar patterns in a code base of billions of lines of code, pinpointing other code segments with similar potential optimization opportunities. Using a fine-tuned LLM, ECO then automatically refactors the code to generate and apply similar edits. Next, ECO verifies the transformed code, submits it for code review, and measures the impact of the optimization in production. Currently deployed on Google's hyperscale production fleet, this system has driven >25k changed lines of production code, across over 6.4k submitted commits, with a >99.5% production success rate. Over the past year, ECO has consistently resulted in significant performance savings every quarter. On average, the savings produced per quarter are equivalent to over 500k normalized CPU cores.

cs.SE↗

arXiv2025

A Bring-Your-Own-Model Approach for ML-Driven Storage Placement in Warehouse-Scale Computers

Storage systems account for a major portion of the total cost of ownership (TCO) of warehouse-scale computers, and thus have a major impact on the overall system's efficiency. Machine learning (ML)-based methods for solving key problems in storage system efficiency, such as data placement, have shown significant promise. However, there are few known practical deployments of such methods. Studying this problem in the context of real-world hyperscale data centers at Google, we identify a number of challenges that we believe cause this lack of practical adoption. Specifically, prior work assumes a monolithic model that resides entirely within the storage layer, an unrealistic assumption in real-world deployments with frequently changing workloads. To address this problem, we introduce a cross-layer approach where workloads instead ''bring their own model''. This strategy moves ML out of the storage system and instead allows each workload to train its own lightweight model at the application layer, capturing the workload's specific characteristics. These small, interpretable models generate predictions that guide a co-designed scheduling heuristic at the storage layer, enabling adaptation to diverse online environments. We build a proof-of-concept of this approach in a production distributed computation framework at Google. Evaluations in a test deployment and large-scale simulation studies using production traces show improvements of as much as 3.47$\times$ in TCO savings compared to state-of-the-art baselines.

cs.DC↗

arXiv2024

LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions

Scheduling virtual machines (VMs) on hosts in cloud data centers dictates efficiency and is an NP-hard problem with incomplete information. Prior work improved VM scheduling with predicted VM lifetimes. Our work further improves lifetime-aware scheduling using repredictions with lifetime distributions versus one-shot prediction. Our approach repredicts and adjusts VM and host lifetimes when incorrect predictions emerge. We also present novel approaches for defragmentation and regular system maintenance, which are essential to our data center reliability and optimizations, and are not explored in prior work. We show repredictions deliver a fundamental advance in effectiveness over one-shot prediction. We call our novel combination of distribution-based lifetime predictions and scheduling algorithms Lifetime Aware VM Allocation (LAVA). LAVA reduces resource stranding and increases the number of empty hosts, which are critical for large VM scheduling, cloud system updates, and reducing dynamic energy consumption. Our approach runs in production within Google's hyperscale cloud data centers, where it improves efficiency by decreasing stranded compute and memory resources by ~3% and ~2% respectively. It increases empty hosts by 2.3-9.2 pp in production, reducing dynamic energy consumption, and increasing availability for large VMs and cloud system updates. We also show a reduction in VM migrations for host defragmentation and maintenance. In addition to our fleet-wide production deployment, we perform simulation studies to characterize the design space and show that our algorithm significantly outperforms the prior state of the art lifetime-based scheduling approach.

cs.DC↗

arXiv2022

Rethinking Storage Management for Data Processing Pipelines in Cloud Data Centers

Data processing frameworks such as Apache Beam and Apache Spark are used for a wide range of applications, from logs analysis to data preparation for DNN training. It is thus unsurprising that there has been a large amount of work on optimizing these frameworks, including their storage management. The shift to cloud computing requires optimization across all pipelines concurrently running across a cluster. In this paper, we look at one specific instance of this problem: placement of I/O-intensive temporary intermediate data on SSD and HDD. Efficient data placement is challenging since I/O density is usually unknown at the time data needs to be placed. Additionally, external factors such as load variability, job preemption, or job priorities can impact job completion times, which ultimately affect the I/O density of the temporary files in the workload. In this paper, we envision that machine learning can be used to solve this problem. We analyze production logs from Google's data centers for a range of data processing pipelines. Our analysis shows that I/O density may be predictable. This suggests that learning-based strategies, if crafted carefully, could extract predictive features for I/O density of temporary files involved in various transformations, which could be used to improve the efficiency of storage management in data processing pipelines.

cs.DC↗

arXiv2022

Learning to Improve Code Efficiency

Improvements in the performance of computing systems, driven by Moore's Law, have transformed society. As such hardware-driven gains slow down, it becomes even more important for software developers to focus on performance and efficiency during development. While several studies have demonstrated the potential from such improved code efficiency (e.g., 2x better generational improvements compared to hardware), unlocking these gains in practice has been challenging. Reasoning about algorithmic complexity and the interaction of coding patterns on hardware can be challenging for the average programmer, especially when combined with pragmatic constraints around development velocity and multi-person development. This paper seeks to address this problem. We analyze a large competitive programming dataset from the Google Code Jam competition and find that efficient code is indeed rare, with a 2x runtime difference between the median and the 90th percentile of solutions. We propose using machine learning to automatically provide prescriptive feedback in the form of hints, to guide programmers towards writing high-performance code. To automatically learn these hints from the dataset, we propose a novel discrete variational auto-encoder, where each discrete latent variable represents a different learned category of code-edit that increases performance. We show that this method represents the multi-modal space of code efficiency edits better than a sequence-to-sequence baseline and generates a distribution of more efficient solutions.

cs.SE↗

arXiv2022

Distilling the Real Cost of Production Garbage Collectors

Abridged abstract: despite the long history of garbage collection (GC) and its prevalence in modern programming languages, there is surprisingly little clarity about its true cost. Without understanding their cost, crucial tradeoffs made by garbage collectors (GCs) go unnoticed. This can lead to misguided design constraints and evaluation criteria used by GC researchers and users, hindering the development of high-performance, low-cost GCs. In this paper, we develop a methodology that allows us to empirically estimate the cost of GC for any given set of metrics. By distilling out the explicitly identifiable GC cost, we estimate the intrinsic application execution cost using different GCs. The minimum distilled cost forms a baseline. Subtracting this baseline from the total execution costs, we can then place an empirical lower bound on the absolute costs of different GCs. Using this methodology, we study five production GCs in OpenJDK 17, a high-performance Java runtime. We measure the cost of these collectors, and expose their respective key performance tradeoffs. We find that with a modestly sized heap, production GCs incur substantial overheads across a diverse suite of modern benchmarks, spending at least 7-82% more wall-clock time and 6-92% more CPU cycles relative to the baseline cost. We show that these costs can be masked by concurrency and generous provisioning of memory/compute. In addition, we find that newer low-pause GCs are significantly more expensive than older GCs, and, surprisingly, sometimes deliver worse application latency than stop-the-world GCs. Our findings reaffirm that GC is by no means a solved problem and that a low-cost, low-latency GC remains elusive. We recommend adopting the distillation methodology together with a wider range of cost metrics for future GC evaluations.

cs.PF↗

arXiv2022

Analyzing a Caching Model

Machine Learning has been successfully applied in systems applications such as memory prefetching and caching, where learned models have been shown to outperform heuristics. However, the lack of understanding the inner workings of these models -- interpretability -- remains a major obstacle for adoption in real-world deployments. Understanding a model's behavior can help system administrators and developers gain confidence in the model, understand risks, and debug unexpected behavior in production. Interpretability for models used in computer systems poses a particular challenge: Unlike ML models trained on images or text, the input domain (e.g., memory access patterns, program counters) is not immediately interpretable. A major challenge is therefore to explain the model in terms of concepts that are approachable to a human practitioner. By analyzing a state-of-the-art caching model, we provide evidence that the model has learned concepts beyond simple statistics that can be leveraged for explanations. Our work provides a first step towards explanability of system ML models and highlights both promises and challenges of this emerging research area.

cs.LG↗

arXiv2019

Boundary-element method to analyze acoustic scattering from a coupled swimbladder-fish body configuration

A model for computing acoustic scattering by a swimbladdered fish with coupling to surrounding fish tissue that is assumed to behave as a homogeneous fluid, is presented. Mathematically, this corresponds to considering the problem of two penetrable scatterers immersed in a homogeneous medium, one of which is wholly embedded in the other. The model is formulated in the frame of boundary integral equations whose solution is achieved using the Boundary Element Method (BEM) for a planar triangle mesh. The numerical implementation is verified against benchmark solutions reported in the literature. The model is then applied to a specimen of \textit{Merluccius hubbsi}, whose morphometry was determined by CT scanning, for evaluating its forward and backscattering responses. From the acoustic scattering viewpoint, the swimbladder is considered as a gas-filled object while the flesh constituting the fish body acts like a weak scatterer. The numerical results suggest the swimbladder and the fish body responses, when fully coupled, can lead to substantial differences with respect to the simplified models normally in use in the area of aquatic ecosystem research.

physics.comp-ph↗