Matthew Lentz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Matthew Lentz has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Danyang Zhuo
Duke University
5 shared publications - Yongji Wu
Duke University
3 shared publications - Beidi Chen
Carnegie Mellon University
2 shared publications - Ceyu Xu
The Hong Kong University of Science and Technology, Hong Kong, Hong Kong
2 shared publications - Lisa Wu Wills
Duke University, Durham, USA
2 shared publications - Xinyu Yang
Carnegie Mellon University
2 shared publications
Latest Publications
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 20-28
The Rise of Tensor Codecs
Video codecs have been refined for decades to compress visual data efficiently. This paper revisits them as general-purpose tensor codecs for large language models. LLM training and inference are increasingly limited by memory footprint and communication bandwidth, while existing tensor-compression methods are often specialized for weights, activations, gradients, or KV cache. We show that standard video codecs can compress these tensors effectively because tensor layouts expose statistical structure that codec stages such as prediction, transform coding, quantization, and entropy coding can exploit. Building on this observation, LLM.265 repurposes existing GPU video engines to support compression during both inference and distributed training. We further show that video-codec hardware can be customized and enhanced to improve tensor encoding and decoding throughput without substantial cost, making it an effective path for large-scale model deployment without major changes to GPU architecture.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 20-28
The Rise of Tensor Codecs
Video codecs have been refined for decades to compress visual data efficiently. This paper revisits them as general-purpose tensor codecs for large language models. LLM training and inference are increasingly limited by memory footprint and communication bandwidth, while existing tensor-compression methods are often specialized for weights, activations, gradients, or KV cache. We show that standard video codecs can compress these tensors effectively because tensor layouts expose statistical structure that codec stages such as prediction, transform coding, quantization, and entropy coding can exploit. Building on this observation, LLM.265 repurposes existing GPU video engines to support compression during both inference and distributed training. We further show that video-codec hardware can be customized and enhanced to improve tensor encoding and decoding throughput without substantial cost, making it an effective path for large-scale model deployment without major changes to GPU architecture.
- Open Access
- 2 April 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 4 (1) , 1-26
HONEYBEE: Efficient Role-based Access Control for Vector Databases via Dynamic Partitioning
Enterprise deployments of vector databases require access control policies to protect sensitive data. These systems often implement access control through hybrid vector queries that combine nearest-neighbor search with relational predicates based on user permissions. However, existing approaches face a fundamental trade-off: dedicated per-user indexes minimize query latency but incur high memory redundancy, while shared indexes with post-search filtering reduce memory overhead at the cost of increased latency. This paper introduces HONEYBEE, a dynamic partitioning framework that leverages the structure of Role-Based Access Control (RBAC) policies to create a smooth trade-off between these extremes. RBAC policies organize users into roles and assign permissions at the role level, 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 memory overhead. To guide these decisions, HONEYBEE develops analytical models of vector search performance and recall, and formulates partitioning as a constrained optimization problem that balances memory usage, query efficiency, and recall. Evaluations on RBAC workloads demonstrate that HONEYBEE achieves up to 13.5X lower query latency than row-level security with only a 1.24X increase in memory usage, while achieving comparable query performance to dedicated, per-role indexes with 90.4% reduction in additional memory consumption, offering a practical middle ground for secure and efficient vector search.
- 2 April 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 4 (1) , 1-26
Open AccessHONEYBEE: Efficient Role-based Access Control for Vector Databases via Dynamic Partitioning
Enterprise deployments of vector databases require access control policies to protect sensitive data. These systems often implement access control through hybrid vector queries that combine nearest-neighbor search with relational predicates based on user permissions. However, existing approaches face a fundamental trade-off: dedicated per-user indexes minimize query latency but incur high memory redundancy, while shared indexes with post-search filtering reduce memory overhead at the cost of increased latency. This paper introduces HONEYBEE, a dynamic partitioning framework that leverages the structure of Role-Based Access Control (RBAC) policies to create a smooth trade-off between these extremes. RBAC policies organize users into roles and assign permissions at the role level, 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 memory overhead. To guide these decisions, HONEYBEE develops analytical models of vector search performance and recall, and formulates partitioning as a constrained optimization problem that balances memory usage, query efficiency, and recall. Evaluations on RBAC workloads demonstrate that HONEYBEE achieves up to 13.5X lower query latency than row-level security with only a 1.24X increase in memory usage, while achieving comparable query performance to dedicated, per-role indexes with 90.4% reduction in additional memory consumption, offering a practical middle ground for secure and efficient vector search.
- Open Access
- 17 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 445-460
LLM.265: Video Codecs are Secretly Tensor Codecs
- 17 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 445-460
Open AccessLLM.265: Video Codecs are Secretly Tensor Codecs
- Open Access
- 14 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-41
Can Large Language Models Verify System Software? A Case Study Using FSCQ as a Benchmark
- 14 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-41
Open AccessCan Large Language Models Verify System Software? A Case Study Using FSCQ as a Benchmark
- Open Access
- 4 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 679-690
MCCS: A Service-based Approach to Collective Communication for Multi-Tenant Cloud
Performance of collective communication is critical for distributed systems. Using libraries to implement collective communication algorithms is not a good fit for a multi-tenant cloud environment because the tenant is not aware of the underlying physical network configuration or how other tenants use the shared cloud network---this lack of information prevents the library from selecting an optimal algorithm. In this paper, we explore a new approach for collective communication that more tightly integrates the implementation with the cloud network instead of the applications. We introduce MCCS, or Managed Collective Communication as a Service, which exposes traditional collective communication abstractions to applications while providing control and flexibility to the cloud provider for their implementations. Realizing MCCS involves overcoming several key challenges to integrate collective communication as part of the cloud network, including memory management of tenant GPU buffers, synchronizing changes to collective communication strategies, and supporting policies that involve cross-layer traffic optimization. Our evaluations show that MCCS improves tenant collective communication performance by up to 2.4× compared to one of the state-of-the-art collective communication libraries (NCCL), while adding more management features including dynamic algorithm adjustment, quality of service, and network-aware traffic engineering.
- 4 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 679-690
Open AccessMCCS: A Service-based Approach to Collective Communication for Multi-Tenant Cloud
Performance of collective communication is critical for distributed systems. Using libraries to implement collective communication algorithms is not a good fit for a multi-tenant cloud environment because the tenant is not aware of the underlying physical network configuration or how other tenants use the shared cloud network---this lack of information prevents the library from selecting an optimal algorithm. In this paper, we explore a new approach for collective communication that more tightly integrates the implementation with the cloud network instead of the applications. We introduce MCCS, or Managed Collective Communication as a Service, which exposes traditional collective communication abstractions to applications while providing control and flexibility to the cloud provider for their implementations. Realizing MCCS involves overcoming several key challenges to integrate collective communication as part of the cloud network, including memory management of tenant GPU buffers, synchronizing changes to collective communication strategies, and supporting policies that involve cross-layer traffic optimization. Our evaluations show that MCCS improves tenant collective communication performance by up to 2.4× compared to one of the state-of-the-art collective communication libraries (NCCL), while adding more management features including dynamic algorithm adjustment, quality of service, and network-aware traffic engineering.
- Open Access
- 30 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 142-157
Dissecting Overheads of Service Mesh Sidecars
- 30 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 142-157
Open AccessDissecting Overheads of Service Mesh Sidecars
- Open Access
- 16 November 2020
- Published by Association for Computing Machinery (ACM)
- p. 774-775
Poirot: private contact summary aggregation
Physical distancing between individuals is key to preventing the spread of a disease such as COVID-19. On the one hand, having access to information about physical interactions is critical for decision makers; on the other, this information is sensitive and can be used to track individuals. In this work, we design Poirot, a system to collect aggregate statistics about physical interactions in a privacy-preserving manner. We show a preliminary evaluation of our system that demonstrates the scalability of our approach even while maintaining strong privacy guarantees.
- 16 November 2020
- Published by Association for Computing Machinery (ACM)
- p. 774-775
Open AccessPoirot: private contact summary aggregation
Physical distancing between individuals is key to preventing the spread of a disease such as COVID-19. On the one hand, having access to information about physical interactions is critical for decision makers; on the other, this information is sensitive and can be used to track individuals. In this work, we design Poirot, a system to collect aggregate statistics about physical interactions in a privacy-preserving manner. We show a preliminary evaluation of our system that demonstrates the scalability of our approach even while maintaining strong privacy guarantees.
- Open Access
- 4 October 2015
- Published by Association for Computing Machinery (ACM)
- p. 230-244
Drowsy power management
Portable computing devices have fast multi-core processors, large memories, and many on-board sensors and radio interfaces, but are often limited by their energy consumption. Traditional power management subsystems have been extended for smartphones and other portable devices, with the intention of maximizing the time that the devices are in a low-power \"sleep\" state. The approaches taken by these subsystems prove inefficient for many short-lived tasks common to portable devices, e.g., querying a sensor or polling a cloud service. We introduce Drowsy, a new power management state that replaces \"awake.\" In the Drowsy state, not all system components are woken up, only the minimal set required for a pending task(s). Drowsy constructs and maintains the minimal task set by dynamically and continuously inferring dependencies between system components at run-time. We have implemented Drowsy within Android, and our results show a significant improvement (1.5-5x) in energy efficiency for common short-lived tasks.
- 4 October 2015
- Published by Association for Computing Machinery (ACM)
- p. 230-244
Open AccessDrowsy power management
Portable computing devices have fast multi-core processors, large memories, and many on-board sensors and radio interfaces, but are often limited by their energy consumption. Traditional power management subsystems have been extended for smartphones and other portable devices, with the intention of maximizing the time that the devices are in a low-power \"sleep\" state. The approaches taken by these subsystems prove inefficient for many short-lived tasks common to portable devices, e.g., querying a sensor or polling a cloud service. We introduce Drowsy, a new power management state that replaces \"awake.\" In the Drowsy state, not all system components are woken up, only the minimal set required for a pending task(s). Drowsy constructs and maintains the minimal task set by dynamically and continuously inferring dependencies between system components at run-time. We have implemented Drowsy within Android, and our results show a significant improvement (1.5-5x) in energy efficiency for common short-lived tasks.