arXiv subjects

Adam Belay

Publications and source records attributed to Adam Belay.

9 recordsLinked to original sources

arXiv2026

Triton for MTIA: Bridging the Programming Model Gaps for Custom AI Accelerators

The rapid growth in machine learning workloads has fueled the proliferation of custom accelerator architectures. Designed from the ground up, these accelerators often expose programming models that are distinct from GPUs. While hyperscalers and AI chip startups continue to innovate in this space, achieving broad operator coverage to support diverse models remains a major challenge. Additionally, an easy-to-use, high-level kernel programming language is important for rapid iteration of models and kernels. Triton, together with TorchInductor, addresses these issues on GPUs, but its viability on accelerators with different programming models has yet to be established. In this work, we present the first production-scale application of Triton on a custom ML accelerator, MTIA-2i, developed by Meta. To support MTIA-2i, we develop a new compiler backend that targets it, introduce enhancements to TorchInductor code generation, and propose minimal language extensions that expose MTIA-specific architectural features. We demonstrate that Triton-MTIA kernels achieve performance competitive with expert-tuned C++ implementations. Leveraging these development efficiency gains, we successfully deployed manually written and Inductor-generated Triton kernels in production across approximately 60 different model types, accounting for 50% of layers and 47% of non-GEMM execution time for these models. Our results provide compelling evidence that DSLs like Triton can bridge the programming model gaps between ML frameworks, kernels, and custom accelerators, enabling rapid innovation and efficient deployment at scale.

cs.PL↗

arXiv2026

Benchmarking Compound AI Applications for Hardware-Software Co-Design

Compound AI applications, composed from interactions between Large Language Models (LLMs), Machine Learning (ML) models, external tools and data sources are quickly becoming an integral workload in datacenters. Their diverse sub-components and use-cases present a large configuration-space across the deployment stack -- ranging from applications and serving software down to hardware -- each of which may influence the application performance, deployment cost, and/or resource consumption. Despite their rapid adoption, however, the systems community lacks a standardized benchmark for analyzing this complicated design-space and guiding in system design. In this work, we present our benchmarking suite used for cross-stack analysis of Compound AI applications. Using this, we derive key takeaways and design principles spanning several layers of the stack for hardware-software co-design to unlock higher resource-efficiency.

cs.DC↗

arXiv2025

Taming Serverless Cold Starts Through OS Co-Design

Serverless computing promises fine-grained elasticity and operational simplicity, fueling widespread interest from both industry and academia. Yet this promise is undercut by the cold setart problem, where invoking a function after a period of inactivity triggers costly initialization before any work can begin. Even with today's high-speed storage, the prevailing view is that achieving sub-millisecond cold starts requires keeping state resident in memory. This paper challenges that assumption. Our analysis of existing snapshot/restore mechanisms show that OS-level limitations, not storage speed, are the real barrier to ultra-fast restores from disk. These limitations force current systems to either restore state piecemeal in a costly manner or capture too much state, leading to longer restore times and unpredictable performance. Futhermore, current memory primitives exposed by the OS make it difficult to reliably fetch data into memory and avoid costly runtime page faults. To overcome these barriers, we present Spice, an execution engine purpose-built for serverless snapshot/restore. Spice integrates directly with the OS to restore kernel state without costly replay and introduces dedicated primitives for restoring memory mappings efficiently and reliably. As a result, Spice delivers near-warm performance on cold restores from disk, reducing latency by up to 14.9x over state-of-the-art process-based systems and 10.6x over VM-based systems. This proves that high performance and memory elasticity no longer need to be a trade-off in serverless computing.

cs.OS↗

arXiv2025

Murakkab: Resource-Efficient Agentic Workflow Orchestration in Cloud Platforms

Agentic workflows commonly coordinate multiple models and tools with complex control logic. They are quickly becoming the dominant paradigm for AI applications. However, serving them remains inefficient with today's frameworks. The key problem is that they expose workflows as opaque sequences of model and tool calls that tightly couple agent logic with model and hardware choices. Often, these workflow components are fragmented across different entities, preventing systems from reasoning about trade-offs across accuracy, latency, energy, and cost. This leads to resource waste and degraded service-level objectives (SLOs). We present Murakkab, a resource-efficient serving system for agentic workflows. Murakkab introduces a declarative abstraction that decouples workflow specification from execution configuration. A profile-guided optimizer and adaptive runtime jointly manage the full stack: orchestrating workflow components, mapping them to models and hardware, and dynamically reconfiguring execution to satisfy user-defined SLOs. By exposing the internal structure of agentic workflows, Murakkab enables cross-layer optimization that existing frameworks and cloud schedulers cannot achieve. Our evaluation on diverse workflows shows that Murakkab reduces GPU usage by up to 2.8$\times$, energy consumption by 3.7$\times$, and cost by 4.3$\times$ while maintaining SLOs.

cs.MA↗

arXiv2025

Checkmate: Zero-Overhead Model Checkpointing via Network Gradient Replication

This paper presents Checkmate, a system that enables per-iteration checkpointing in DNN training without any training slowdown. The traditional approach to checkpointing requires a pause in training to copy model states to a separate location, allowing the state to be restored in the event of failure. This approach fundamentally has a tradeoff between the frequency of checkpoints and the cost of a failure. We avoid this tradeoff; our key insight is that in data-parallel training, all information necessary to create a checkpoint already exists in the network as gradients. Our core contribution is a new multicast abstraction that simultaneously delivers gradients to a separate CPU-based shadow cluster. The shadow maintains a checkpoint by applying those gradients to a copy of the model. Our evaluation shows that Checkmate performs per-iteration checkpointing with training throughput comparable to an ideal no-checkpoint baseline. Checkmate achieves 5 to 34.5x more frequent checkpointing compared to state-of-the-art checkpointing systems, resulting in 80% to 97.1% reduction in repeated work per failure. At the same checkpointing frequency, Checkmate delivers 1.3x to 6.5x throughput compared to other systems.

cs.DC↗

arXiv2025

Towards Resource-Efficient Compound AI Systems

Compound AI Systems, integrating multiple interacting components like models, retrievers, and external tools, have emerged as essential for addressing complex AI tasks. However, current implementations suffer from inefficient resource utilization due to tight coupling between application logic and execution details, a disconnect between orchestration and resource management layers, and the perceived exclusiveness between efficiency and quality. We propose a vision for resource-efficient Compound AI Systems through a declarative workflow programming model and an adaptive runtime system for dynamic scheduling and resource-aware decision-making. Decoupling application logic from low-level details exposes levers for the runtime to flexibly configure the execution environment and resources, without compromising on quality. Enabling collaboration between the workflow orchestration and cluster manager enables higher efficiency through better scheduling and resource management. We are building a prototype system, called Murakkab, to realize this vision. Our preliminary evaluation demonstrates speedups up to $\sim 3.4\times$ in workflow completion times while delivering $\sim 4.5\times$ higher energy efficiency, showing promise in optimizing resources and advancing AI system design.

cs.DC↗

arXiv2024

Junctiond: Extending FaaS Runtimes with Kernel-Bypass

This report explores the use of kernel-bypass networking in FaaS runtimes and demonstrates how using Junction, a novel kernel-bypass system, as the backend for executing components in faasd can enhance performance and isolation. Junction achieves this by reducing network and compute overheads and minimizing interactions with the host operating system. Junctiond, the integration of Junction with faasd, reduces median and P99 latency by 37.33% and 63.42%, respectively, and can handle 10 times more throughput while decreasing latency by 2x at the median and 3.5 times at the tail.

cs.DC↗

arXiv2022

Treehouse: A Case For Carbon-Aware Datacenter Software

The end of Dennard scaling and the slowing of Moore's Law has put the energy use of datacenters on an unsustainable path. Datacenters are already a significant fraction of worldwide electricity use, with application demand scaling at a rapid rate. We argue that substantial reductions in the carbon intensity of datacenter computing are possible with a software-centric approach: by making energy and carbon visible to application developers on a fine-grained basis, by modifying system APIs to make it possible to make informed trade offs between performance and carbon emissions, and by raising the level of application programming to allow for flexible use of more energy efficient means of compute and storage. We also lay out a research agenda for systems software to reduce the carbon footprint of datacenter computing.

cs.DC↗

arXiv2021

Efficient Strong Scaling Through Burst Parallel Training

As emerging deep neural network (DNN) models continue to grow in size, using large GPU clusters to train DNNs is becoming an essential requirement to achieving acceptable training times. In this paper, we consider the case where future increases in cluster size will cause the global batch size that can be used to train models to reach a fundamental limit: beyond a certain point, larger global batch sizes cause sample efficiency to degrade, increasing overall time to accuracy. As a result, to achieve further improvements in training performance, we must instead consider "strong scaling" strategies that hold the global batch size constant and allocate smaller batches to each GPU. Unfortunately, this makes it significantly more difficult to use cluster resources efficiently. We present DeepPool, a system that addresses this efficiency challenge through two key ideas. First, burst parallelism allocates large numbers of GPUs to foreground jobs in bursts to exploit the unevenness in parallelism across layers. Second, GPU multiplexing prioritizes throughput for foreground training jobs, while packing in background training jobs to reclaim underutilized GPU resources, thereby improving cluster-wide utilization. Together, these two ideas enable DeepPool to deliver a 1.2 - 2.3x improvement in total cluster throughput over standard data parallelism with a single task when the cluster scale is large.

cs.DC↗