[Paper Review] Sieve: Actionable Insights from Monitored Metrics in Microservices

Jörg Thalheim, António Rodrigues+6 more|arXiv (Cornell University)|Sep 20, 2017

Software System Performance and Reliability48 references11 citations

TL;DR

Sieve is a platform that derives actionable insights from monitored metrics in microservices by reducing metric dimensionality and inferring causal dependencies using Granger Causality. It reduces metrics by 10–100×, cuts system overhead by up to 90% in storage and 80% in CPU, and enables effective autoscaling and root cause analysis without instrumentation or prior knowledge.

ABSTRACT

Major cloud computing operators provide powerful monitoring tools to understand the current (and prior) state of the distributed systems deployed in their infrastructure. While such tools provide a detailed monitoring mechanism at scale, they also pose a significant challenge for the application developers/operators to transform the huge space of monitored metrics into useful insights. These insights are essential to build effective management tools for improving the efficiency, resiliency, and dependability of distributed systems. This paper reports on our experience with building and deploying Sieve - a platform to derive actionable insights from monitored metrics in distributed systems. Sieve builds on two core components: a metrics reduction framework, and a metrics dependency extractor. More specifically, Sieve first reduces the dimensionality of metrics by automatically filtering out unimportant metrics by observing their signal over time. Afterwards, Sieve infers metrics dependencies between distributed components of the system using a predictive-causality model by testing for Granger Causality. We implemented Sieve as a generic platform and deployed it for two microservices-based distributed systems: OpenStack and ShareLatex. Our experience shows that (1) Sieve can reduce the number of metrics by at least an order of magnitude (10 - 100$ imes$), while preserving the statistical equivalence to the total number of monitored metrics; (2) Sieve can dramatically improve existing monitoring infrastructures by reducing the associated overheads over the entire system stack (CPU - 80%, storage - 90%, and network - 50%); (3) Lastly, Sieve can be effective to support a wide-range of workflows in distributed systems - we showcase two such workflows: orchestration of autoscaling, and Root Cause Analysis (RCA).

Motivation & Objective

• To address the challenge of transforming high-dimensional, monitored metrics into actionable insights for microservices management.
• To reduce the number of monitored metrics without losing statistical relevance, minimizing monitoring overhead.
• To infer causal relationships between metrics across distributed components to support system management workflows.
• To enable unsupervised, generic, and application-agnostic insight derivation from existing monitoring data.
• To support critical DevOps workflows such as autoscaling and root cause analysis using inferred metric dependencies.

Proposed method

• Sieve employs a metrics reduction engine that filters out unimportant metrics by analyzing their temporal signal strength and variance.
• It applies a predictive-causality model based on Granger Causality to infer causal relationships between metrics of different components.
• The system operates in an unsupervised manner, requiring no prior knowledge of metric time series or application semantics.
• It uses workload generators or production traces to validate the inferred causal models across real-world microservices.
• The platform is deployed on OpenStack and ShareLatex, demonstrating scalability and adaptability across diverse systems.
• Sieve integrates with existing monitoring stacks without requiring instrumentation or changes to the monitored system.

Experimental results

Research questions

• RQ1 How can we automatically reduce the dimensionality of monitored metrics in microservices while preserving statistical relevance?
• RQ2 Can we infer meaningful causal dependencies between metrics across distributed components using only time-series data?
• RQ3 To what extent can metric reduction and dependency inference reduce monitoring overhead in real-world systems?
• RQ4 Can the inferred metric dependencies effectively support real-world management workflows like autoscaling and root cause analysis?
• RQ5 How can such a system be deployed in a generic, unsupervised, and non-intrusive manner across diverse microservices architectures?

Key findings

• Sieve reduces the number of monitored metrics by at least 10× to 100× while preserving statistical equivalence to the full metric set.
• The system reduces CPU overhead by up to 80%, storage by up to 90%, and network usage by 50% across the entire system stack.
• The inferred causal models enable effective autoscaling by identifying key metrics that drive resource demand.
• The platform supports root cause analysis by identifying the earliest metrics that deviate, enabling faster failure diagnosis.
• Sieve’s approach is generic and reusable across different microservices systems, as demonstrated on OpenStack and ShareLatex.
• The system operates without instrumentation or prior knowledge, relying solely on observed metric time series.