Matthew Lentz

Yongji Wu, Matthew Lentz, Danyang Zhuo et al.

With the advent of ubiquitous deployment of smart devices and the Internet of Things, data sources for machine learning inference have increasingly moved to the edge of the network. Existing machine learning inference platforms typically assume a homogeneous infrastructure and do not take into account the more complex and tiered computing infrastructure that includes edge devices, local hubs, edge datacenters, and cloud datacenters. On the other hand, recent AutoML efforts have provided viable solutions for model compression, pruning and quantization for heterogeneous environments; for a machine learning model, now we may easily find or even generate a series of models with different tradeoffs between accuracy and efficiency. We design and implement JellyBean, a system for serving and optimizing machine learning inference workflows on heterogeneous infrastructures. Given service-level objectives (e.g., throughput, accuracy), JellyBean picks the most cost-efficient models that meet the accuracy target and decides how to deploy them across different tiers of infrastructures. Evaluations show that JellyBean reduces the total serving cost of visual question answering by up to 58%, and vehicle tracking from the NVIDIA AI City Challenge by up to 36% compared with state-of-the-art model selection and worker assignment solutions. JellyBean also outperforms prior ML serving systems (e.g., Spark on the cloud) up to 5x in serving costs.

Yongji Wu, Wenjie Qu, Xueshen Liu et al.

Sparsely-activated Mixture-of-Experts (MoE) architecture has increasingly been adopted to further scale large language models (LLMs). However, frequent failures still pose significant challenges as training scales. The cost of even a single failure is significant, as all GPUs need to idle wait until the failure is resolved, potentially losing considerable training progress as training has to restart from checkpoints. This problem is exacerbated by the growing use of spot instances on public clouds for model training, which despite offering substantial cost savings, introduce frequent preemptions-essentially failures that regularly occur throughout the training process. Existing solutions for efficient fault-tolerant training either lack elasticity or rely on building resiliency into pipeline parallelism, which cannot be applied to MoE models due to the expert parallelism strategy adopted by the MoE architecture. We present Lazarus, a system for resilient and elastic training of MoE models. Lazarus adaptively allocates expert replicas to address the inherent imbalance in expert workload and speeds up training, while a provably optimal expert placement algorithm is developed to maximize the probability of recovery upon failures. Through adaptive expert placement and a flexible token dispatcher, Lazarus can also fully utilize all available nodes after failures, leaving no GPU idle. Our evaluation shows that Lazarus outperforms existing MoE training systems by up to 5.7x under frequent node failures and 3.4x on a real spot instance trace.

Alexander Du, Jianjun Ou, Danyang Zhuo et al.

Large language models are increasingly used for code generation, but many generated programs fail to compile, a prerequisite for further correctness checks such as unit tests. Existing solutions for repairing static errors are costly in both latency and token consumption. Post-hoc repair delays error detection until generation completes and commonly regenerates large regions of previously valid code. Constrained semantic decoding checks after each token, incurring per-token overhead while limiting repair to the current token even when the root cause lies earlier. We present Hydra, a system for efficient recovery from static errors during code generation. Hydra allows checking to proceed asynchronously with generation, avoiding checker overhead when the generated code is semantically correct. In addition, it provides checkpoint-and-rollback support for targeted repair, avoiding regeneration and rechecking of valid prefixes. We retrofit the Clang C/C++ compiler to support Hydra with modest modifications. Paired with a token-efficient repair strategy, Hydra reduces latency by up to 71% and token consumption by up to 70% relative to post-hoc repair on C/C++ code generation tasks that encounter static errors.

Yao Lu, Song Bian, Lequn Chen et al.

In this paper, we investigate the intersection of large generative AI models and cloud-native computing architectures. Recent large models such as ChatGPT, while revolutionary in their capabilities, face challenges like escalating costs and demand for high-end GPUs. Drawing analogies between large-model-as-a-service (LMaaS) and cloud database-as-a-service (DBaaS), we describe an AI-native computing paradigm that harnesses the power of both cloud-native technologies (e.g., multi-tenancy and serverless computing) and advanced machine learning runtime (e.g., batched LoRA inference). These joint efforts aim to optimize costs-of-goods-sold (COGS) and improve resource accessibility. The journey of merging these two domains is just at the beginning and we hope to stimulate future research and development in this area.

Shuowei Jin, Xueshen Liu, Yongji Wu et al.

Large language models (LLMs) have achieved remarkable success in natural language tasks, but their inference incurs substantial computational and memory overhead. To improve efficiency, parallel decoding methods like Skeleton-of-Thought (SoT) decompose prompts into sub-problems for concurrent processing. However, these methods significantly compromise answer quality by treating semantically linked sub-problems as independent. We propose Plato, a novel approach that co-designs algorithms and systems for semantic-aware parallel decoding. Plato leverages LLMs to organize sub-problems into a dependency graph based on logical and causal relationships, enabling concurrent decoding of non-dependent nodes while preserving answer coherence and quality. To further enhance efficiency, Plato pipelines planning and node decoding stages, implements a global context cache, and carefully structures node inference prompts to maximize key-value cache reuse and minimize overhead. Our evaluations show that Plato improves throughput by 68% over autoregressive decoding while achieving a 40% net win rate in answer quality. Compared to SoT, Plato demonstrates a remarkable 90% quality net-win rate. Ablation studies reveal that our pipeline design improves speedup by 29%, while our KV cache reuse optimization reduces overhead by 75%.

Hongbin Zhong, Matthew Lentz, Nina Narodytska et al.

As vector databases gain traction in enterprise applications, robust access control has become critical to safeguard sensitive data. Access control in these systems is often implemented through hybrid vector queries, which combine nearest neighbor search on vector data with relational predicates based on user permissions. However, existing approaches face significant trade-offs: creating dedicated indexes for each user minimizes query latency but introduces excessive storage redundancy, while building a single index and applying access control after vector search reduces storage overhead but suffers from poor recall and increased query latency. This paper introduces HoneyBee, a dynamic partitioning framework that bridges the gap between these approaches by leveraging the structure of Role-Based Access Control (RBAC) policies. RBAC, widely adopted in enterprise settings, groups users into roles and assigns permissions to those roles, creating a natural "thin waist" in the permission structure that is ideal for partitioning decisions. Specifically, HoneyBee produces overlapping partitions where vectors can be strategically replicated across different partitions to reduce query latency while controlling storage overhead. By introducing analytical models for the performance and recall of the vector search, HoneyBee formulates the partitioning strategy as a constrained optimization problem to dynamically balance storage, query efficiency, and recall. Evaluations on RBAC workloads demonstrate that HoneyBee reduces storage redundancy compared to role partitioning and achieves up to 6x faster query speeds than row-level security (RLS) with only 1.4x storage increase, offering a practical middle ground for secure and efficient vector search.

Yongji Wu, Xueshen Liu, Shuowei Jin et al.

The Mixture-of-Experts (MoE) architecture has become increasingly popular as a method to scale up large language models (LLMs). To save costs, heterogeneity-aware training solutions have been proposed to utilize GPU clusters made up of both newer and older-generation GPUs. However, existing solutions are agnostic to the performance characteristics of different MoE model components (i.e., attention and expert) and do not fully utilize each GPU's compute capability. In this paper, we introduce HeterMoE, a system to efficiently train MoE models on heterogeneous GPUs. Our key insight is that newer GPUs significantly outperform older generations on attention due to architectural advancements, while older GPUs are still relatively efficient for experts. HeterMoE disaggregates attention and expert computation, where older GPUs are only assigned with expert modules. Through the proposed zebra parallelism, HeterMoE overlaps the computation on different GPUs, in addition to employing an asymmetric expert assignment strategy for fine-grained load balancing to minimize GPU idle time. Our evaluation shows that HeterMoE achieves up to 2.3x speed-up compared to existing MoE training systems, and 1.4x compared to an optimally balanced heterogeneity-aware solution. HeterMoE efficiently utilizes older GPUs by maintaining 95% training throughput on average, even with half of the GPUs in a homogeneous A40 cluster replaced with V100.

Ceyu Xu, Yongji Wu, Xinyu Yang et al.

As the parameter size of large language models (LLMs) continues to expand, the need for a large memory footprint and high communication bandwidth have become significant bottlenecks for the training and inference of LLMs. To mitigate these bottlenecks, various tensor compression techniques have been proposed to reduce the data size, thereby alleviating memory requirements and communication pressure. Our research found that video codecs, despite being originally designed for compressing videos, show excellent efficiency when compressing various types of tensors. We demonstrate that video codecs can be versatile and general-purpose tensor codecs while achieving the state-of-the-art compression efficiency in various tasks. We further make use of the hardware video encoding and decoding module available on GPUs to create a framework capable of both inference and training with video codecs repurposed as tensor codecs. This greatly reduces the requirement for memory capacity and communication bandwidth, enabling training and inference of large models on consumer-grade GPUs.

Pierfrancesco Ingo, Nichole Boufford, Ming Cheng Jiang et al.

Epidemics like the recent COVID-19 require proactive contact tracing and epidemiological analysis to predict and subsequently contain infection transmissions. The proactive measures require large scale data collection, which simultaneously raise concerns regarding users' privacy. Digital contact tracing systems developed in response to COVID-19 either collected extensive data for effective analytics at the cost of users' privacy or collected minimal data for the sake of user privacy but were ineffective in predicting and mitigating the epidemic risks. We present Silmarillion--in preparation for future epidemics--a system that reconciles user's privacy with rich data collection for higher utility. In Silmarillion, user devices record Bluetooth encounters with beacons installed in strategic locations. The beacons further enrich the encounters with geo-location, location type, and environment conditions at the beacon installation site. This enriched information enables detailed scientific analysis of disease parameters as well as more accurate personalized exposure risk notification. At the same time, Silmarillion provides privacy to all participants and non-participants at the same level as that guaranteed in digital and manual contact tracing. We describe the design of Silmarillion and its communication protocols that ensure user privacy and data security. We also evaluate a prototype of Silmarillion built using low-end IoT boards, showing that the power consumption and user latencies are adequately low for a practical deployment. Finally, we briefly report on a small-scale deployment within a university building as a proof-of-concept.

Matthew Lentz, Rijurekha Sen, Peter Druschel et al.

Reliable on-off control of peripherals on smart devices is a key to security and privacy in many scenarios. Journalists want to reliably turn off radios to protect their sources during investigative reporting. Users wish to ensure cameras and microphones are reliably off during private meetings. In this paper, we present SeCloak, an ARM TrustZone-based solution that ensures reliable on-off control of peripherals even when the platform software is compromised. We design a secure kernel that co-exists with software running on mobile devices (e.g., Android and Linux) without requiring any code modifications. An Android prototype demonstrates that mobile peripherals like radios, cameras, and microphones can be controlled reliably with a very small trusted computing base and with minimal performance overhead.