arXiv subjects

Ken Birman

Publications and source records attributed to Ken Birman.

10 recordsLinked to original sources

arXiv2025

Passing the Baton: High Throughput Distributed Disk-Based Vector Search with BatANN

Vector search underpins modern information-retrieval systems, including retrieval-augmented generation (RAG) pipelines and search engines over unstructured text and images. As datasets scale to billions of vectors, disk-based vector search has emerged as a practical solution. However, looking to the future, we must anticipate datasets too large for any single server and throughput demands that exceed the limits of locally attached SSDs. We present BatANN, a distributed disk-based approximate nearest neighbor (ANN) system that retains the logarithmic search efficiency of a single global graph while achieving near-linear throughput scaling in the number of servers. Our core innovation is that when accessing a neighborhood which is stored on another machine, we send the full state of the query to the other machine to continue executing there for improved locality. On 1B-point datasets at 0.95 recall using 10 servers, BatANN achieves 3.5-5.59x of the scatter-gather baseline and 1.44-2.09x the throughput of DistributedANN, respectively, while maintaining mean latency below 3 ms. Moreover, we get these results on standard TCP. To our knowledge, BatANN is the first open-source distributed disk-based vector search system to operate over a single global graph.

cs.DC↗

arXiv2025

Vortex: Hosting ML Inference and Knowledge Retrieval Services With Tight Latency and Throughput Requirements

There is growing interest in deploying ML inference and knowledge retrieval as services that could support both interactive queries by end users and more demanding request flows that arise from AIs integrated into a end-user applications and deployed as agents. Our central premise is that these latter cases will bring service level latency objectives (SLOs). Existing ML serving platforms use batching to optimize for high throughput, exposing them to unpredictable tail latencies. Vortex enables an SLO-first approach. For identical tasks, Vortex's pipelines achieve significantly lower and more stable latencies than TorchServe and Ray Serve over a wide range of workloads, often enabling a given SLO target at more than twice the request rate. When RDMA is available, the Vortex advantage is even more significant.

cs.DB↗

arXiv2024

AI-Based Teat Shape and Skin Condition Prediction for Dairy Management

Dairy owners spend significant effort to keep their animals healthy. There is good reason to hope that technologies such as computer vision and artificial intelligence (AI) could reduce these costs, yet obstacles arise when adapting advanced tools to farming environments. In this work, we adapt AI tools to dairy cow teat localization, teat shape, and teat skin condition classifications. We also curate a data collection and analysis methodology for a Machine Learning (ML) pipeline. The resulting teat shape prediction model achieves a mean Average Precision (mAP) of 0.783, and the teat skin condition model achieves a mean average precision of 0.828. Our work leverages existing ML vision models to facilitate the individualized identification of teat health and skin conditions, applying AI to the dairy management industry.

cs.AI↗

arXiv2024

Compass: A Decentralized Scheduler for Latency-Sensitive ML Workflows

We consider ML query processing in distributed systems where GPU-enabled workers coordinate to execute complex queries: a computing style often seen in applications that interact with users in support of image processing and natural language processing. In such systems, coscheduling of GPU memory management and task placement represents a promising opportunity. We propose Compass, a novel framework that unifies these functions to reduce job latency while using resources efficiently, placing tasks where data dependencies will be satisfied, collocating tasks from the same job (when this will not overload the host or its GPU), and efficiently managing GPU memory. Comparison with other state of the art schedulers shows a significant reduction in completion times while requiring the same amount or even fewer resources. In one case, just half the servers were needed for processing the same workload.

cs.DC↗

arXiv2023

Keep Your Friends Close: Leveraging Affinity Groups to Accelerate AI Inference Workflows

AI inference workflows are typically structured as a pipeline or graph of AI programs triggered by events. As events occur, the AIs perform inference or classification tasks under time pressure to respond or take some action. Standard techniques that reduce latency in other streaming settings (such as caching and optimization-driven scheduling) are of limited value because AI data access patterns (models, databases) change depending on the triggering event: a significant departure from traditional streaming. In this work, we propose a novel affinity grouping mechanism that makes it easier for developers to express application-specific data access correlations, enabling coordinated management of data objects in server clusters hosting streaming inference tasks. Our proposals are thus complementary to other approaches such as caching and scheduling. Experiments confirm the limitations of standard techniques, while showing that the proposed mechanism is able to maintain significantly lower latency as workload and scale-out increase, and yet requires only minor code changes.

cs.DC↗

arXiv2023

Cascade: A Platform for Delay-Sensitive Edge Intelligence

Interactive intelligent computing applications are increasingly prevalent, creating a need for AI/ML platforms optimized to reduce per-event latency while maintaining high throughput and efficient resource management. Yet many intelligent applications run on AI/ML platforms that optimize for high throughput even at the cost of high tail-latency. Cascade is a new AI/ML hosting platform intended to untangle this puzzle. Innovations include a legacy-friendly storage layer that moves data with minimal copying and a "fast path" that collocates data and computation to maximize responsiveness. Our evaluation shows that Cascade reduces latency by orders of magnitude with no loss of throughput.

cs.OS↗

arXiv2021

Spindle: Techniques for Optimizing Atomic Multicast on RDMA

Leveraging one-sided RDMA for applications that replicate small data objects can be surprisingly difficult: such uses amplify any protocol overheads. Spindle is a set of optimization techniques for systematically tackling this class of challenges for atomic multicast over RDMA. These include memory polling optimizations using novel sender and receiver batching techniques, null-message send logic, and improved multi-thread synchronization. We applied Spindle to Derecho, an open-source C++ library for atomic multicast, and obtained significant performance improvements both for the library itself and for an OMG-compliant avionics DDS built over Derecho. Derecho's multicast bandwidth utilization for 10KB messages rose from 1GB/s to 9.7GB/s on a 12.5GB/s network, and it became more robust to delays. Interestingly, although some of our techniques employ batching, latency dropped by nearly two orders of magnitude. Spindle optimizations should also be of value in other RDMA applications limited by the speed of coordination.

cs.DC↗

arXiv2015

Cache Serializability: Reducing Inconsistency in Edge Transactions

Read-only caches are widely used in cloud infrastructures to reduce access latency and load on backend databases. Operators view coherent caches as impractical at genuinely large scale and many client-facing caches are updated in an asynchronous manner with best-effort pipelines. Existing solutions that support cache consistency are inapplicable to this scenario since they require a round trip to the database on every cache transaction. Existing incoherent cache technologies are oblivious to transactional data access, even if the backend database supports transactions. We propose T-Cache, a novel caching policy for read-only transactions in which inconsistency is tolerable (won't cause safety violations) but undesirable (has a cost). T-Cache improves cache consistency despite asynchronous and unreliable communication between the cache and the database. We define cache-serializability, a variant of serializability that is suitable for incoherent caches, and prove that with unbounded resources T-Cache implements this new specification. With limited resources, T-Cache allows the system manager to choose a trade-off between performance and consistency. Our evaluation shows that T-Cache detects many inconsistencies with only nominal overhead. We use synthetic workloads to demonstrate the efficacy of T-Cache when data accesses are clustered and its adaptive reaction to workload changes. With workloads based on the real-world topologies, T-Cache detects 43-70% of the inconsistencies and increases the rate of consistent transactions by 33-58%.

cs.DC↗

arXiv2014

Practical Experience Report: The Performance of Paxos in the Cloud

This experience report presents the results of an extensive performance evaluation conducted using four open-source implementations of Paxos deployed in Amazon's EC2. Paxos is a fundamental algorithm for building fault-tolerant services, at the core of state-machine replication. Implementations of Paxos are currently used in many prototypes and production systems in both academia and industry. Although all protocols surveyed in the paper implement Paxos, they are optimized in a number of different ways, resulting in very different behavior, as we show in the paper. We have considered a variety of configurations and failure-free and faulty executions. In addition to reporting our findings, we propose and assess additional optimizations to existing implementations.

cs.DC↗

arXiv1998

The Design and Architecture of the Microsoft Cluster Service -- A Practical Approach to High-Availability and Scalability

Microsoft Cluster Service (MSCS) extends the Win-dows NT operating system to support high-availability services. The goal is to offer an execution environment where off-the-shelf server applications can continue to operate, even in the presence of node failures. Later ver-sions of MSCS will provide scalability via a node and application management system that allows applications to scale to hundreds of nodes. This paper provides a de-tailed description of the MSCS architecture and the de-sign decisions that have driven the implementation of the service. The paper also describes how some major appli-cations use the MSCS features, and describes features added to make it easier to implement and manage fault-tolerant applications on MSCS.

cs.OS↗