Publications

2025

1. VLDB

   Approximation-First Timeseries Query At Scale

   Zeying Zhu, Jonathan Chamberlain, Kenny Wu, and 2 more authors

   Proc. VLDB Endow., Apr 2025

   Abs DOI

   Timeseries monitoring systems such as Prometheus play a crucial role in gaining observability of the underlying system infrastructure. These systems collect timeseries metrics from various system components and perform monitoring queries over periodic window-based aggregations (i.e., rule queries). However, despite wide adoption, the operational costs and query latency of rule queries remain high. In this paper, we identify major bottlenecks associated with repeated data scans and query computations concerning window overlaps in rule queries, and present PromSketch, an approximation-first query framework as intermediate caches for monitoring systems. It enables low operational costs and query latency, by combining approximate window-based query frameworks and sketch-based precomputation. PromSketch is implemented as a standalone module that can be integrated into Prometheus and VictoriaMetrics, covering 70% of Prometheus’ aggregation over time queries. Our evaluation shows that PromSketch achieves up to a two-order-of-magnitude reduction in query latency over Prometheus and VictoriaMetrics, while lowering operational dollar costs of query processing by three orders of magnitude compared to Prometheus and by at least 4\texttimes compared to VictoriaMetrics with at most 5% average errors across statistics.

2. Rearchitecting Approximation-First Cloud Telemetry

   Zeying Zhu and Zaoxing Liu

   In Proceedings of the ACM SIGCOMM 2025 Posters and Demos, Coimbra, Portugal, Apr 2025

   Abs DOI

   Cloud-native telemetry systems like Prometheus enable observability through rule queries-window- and dimension-based aggregations over time series data. However, these queries incur high operational costs and latency due to raw data ingestion, repeated computations, and expensive storage. In this demonstration, we revisit the potential of approximate analytics to improve efficiency across the telemetry pipeline. We propose a hybrid, sketch-based caching architecture that unifies optimizations across ingestion, query processing, and storage. By supporting approximate representations at each stage, our design achieves significant cost savings while maintaining high accuracy and responsiveness.

2024

1. IMC

   Raising the Level of Abstraction for Sketch-Based Network Telemetry with SketchPlan

   Milind Srivastava, Shao-Tse Hung, Hun Namkung, and 3 more authors

   In Proceedings of the 2024 ACM on Internet Measurement Conference, Madrid, Spain, Apr 2024

   Abs DOI

   While sketch-based network telemetry is attractive, realizing its potential benefits has been elusive in practice. Existing sketch solutions offer low-level interfaces and impose high effort on operators to satisfy telemetry intents with required accuracies. Extending these approaches to reduce effort results in inefficient deployments with poor accuracy-resource tradeoffs. We present SketchPlan, an abstraction layer for sketch-based telemetry to reduce effort and achieve high efficiency. SketchPlan takes an ensemble view across telemetry intents and sketches, instead of existing approaches that consider each intent-sketch pair in isolation. We show that SketchPlan improves accuracy-resource tradeoffs by up-to 12x and up-to 60x vs. baselines, in single-node and network-wide settings. SketchPlan is open-sourced at: https://github.com/milindsrivastava1997/SketchPlan.

2. SIGCOMM

   Poster: Circa: Re-imagining Network Telemetry from an Approximation-First Perspective

   Milind Srivastava and Vyas Sekar

   In Proceedings of the ACM SIGCOMM 2024 Conference: Posters and Demos, Sydney, NSW, Australia, Apr 2024

   Abs DOI

   Telemetry systems are widely used to collect data from distributed endpoints, analyze data in conjunction to gain valuable insights, and store data for historical analytics. These systems consist of four stages (Figure 1): collection, transmission, analysis, and storage. Collectors at the endpoint collect various types of data, which is then transmitted to a central server for analysis. This data is used for multiple downstream tasks, such as dashboard monitoring and anomaly detection. Finally, this data is stored in long-term storage to aid retrospective analytics and debugging.