Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Morphis: SLO-Aware Resource Scheduling for Microservices with Time-Varying Call Graphs

Yu Tang, Hailiang Zhao, Rui Shi, Chuansheng Lu, Yifei Zhang, Kingsum Chow, Shuiguang Deng

TL;DR

Morphis addresses the challenge of dynamic microservice call graphs by uncovering latent, recurrent invocation patterns and decomposing traces into a stable backbone plus deviation subgraphs. It couples structural fingerprinting with a latency-aware resource model and pattern-based load forecasting to drive a global optimization that coordinates per-service latency targets under end-to-end SLO constraints. The framework demonstrates up to $38\%$ CPU savings while maintaining high SLO compliance ($\ge 95\%$) on the TrainTicket benchmark, outperforming reactive and static-baseline autoscalers. This pattern-driven, end-to-end approach enables proactive, coordinated provisioning in evolving microservice topologies, with strong practical implications for cloud-native platforms.

Abstract

Modern microservice systems exhibit continuous structural evolution in their runtime call graphs due to workload fluctuations, fault responses, and deployment activities. Despite this complexity, our analysis of over 500,000 production traces from ByteDance reveals a latent regularity: execution paths concentrate around a small set of recurring invocation patterns. However, existing resource management approaches fail to exploit this structure. Industrial autoscalers like Kubernetes HPA ignore inter-service dependencies, while recent academic methods often assume static topologies, rendering them ineffective under dynamic execution contexts. In this work, we propose Morphis, a dependency-aware provisioning framework that unifies pattern-aware trace analysis with global optimization. It introduces structural fingerprinting that decomposes traces into a stable execution backbone and interpretable deviation subgraphs. Then, resource allocation is formulated as a constrained optimization problem over predicted pattern distributions, jointly minimizing aggregate CPU usage while satisfying end-to-end tail-latency SLOs. Our extensive evaluations on the TrainTicket benchmark demonstrate that Morphis reduces CPU consumption by 35-38% compared to state-of-the-art baselines while maintaining 98.8% SLO compliance.

Morphis: SLO-Aware Resource Scheduling for Microservices with Time-Varying Call Graphs

TL;DR

Morphis addresses the challenge of dynamic microservice call graphs by uncovering latent, recurrent invocation patterns and decomposing traces into a stable backbone plus deviation subgraphs. It couples structural fingerprinting with a latency-aware resource model and pattern-based load forecasting to drive a global optimization that coordinates per-service latency targets under end-to-end SLO constraints. The framework demonstrates up to CPU savings while maintaining high SLO compliance () on the TrainTicket benchmark, outperforming reactive and static-baseline autoscalers. This pattern-driven, end-to-end approach enables proactive, coordinated provisioning in evolving microservice topologies, with strong practical implications for cloud-native platforms.

Abstract

Modern microservice systems exhibit continuous structural evolution in their runtime call graphs due to workload fluctuations, fault responses, and deployment activities. Despite this complexity, our analysis of over 500,000 production traces from ByteDance reveals a latent regularity: execution paths concentrate around a small set of recurring invocation patterns. However, existing resource management approaches fail to exploit this structure. Industrial autoscalers like Kubernetes HPA ignore inter-service dependencies, while recent academic methods often assume static topologies, rendering them ineffective under dynamic execution contexts. In this work, we propose Morphis, a dependency-aware provisioning framework that unifies pattern-aware trace analysis with global optimization. It introduces structural fingerprinting that decomposes traces into a stable execution backbone and interpretable deviation subgraphs. Then, resource allocation is formulated as a constrained optimization problem over predicted pattern distributions, jointly minimizing aggregate CPU usage while satisfying end-to-end tail-latency SLOs. Our extensive evaluations on the TrainTicket benchmark demonstrate that Morphis reduces CPU consumption by 35-38% compared to state-of-the-art baselines while maintaining 98.8% SLO compliance.
Paper Structure (29 sections, 16 equations, 13 figures, 1 table, 2 algorithms)

This paper contains 29 sections, 16 equations, 13 figures, 1 table, 2 algorithms.

Table of Contents

  1. Background and Motivation
  2. Dynamism with Latent Regularity
  3. The Case for Dependency-Aware Resource Management
  4. The Morphis Framework
  5. Structural Fingerprinting
  6. Problem Formulation
  7. Backbone Extraction
  8. Deviation Subgraph Extraction
  9. Similarity Computation
  10. Resource Modeling and Optimization
  11. Latency-Aware Resource Modeling
  12. Pattern-Based Load Forecasting
  13. Global Resource Allocation Optimization
  14. Evaluation
  15. System Implementation
  16. ...and 14 more sections

Figures (13)

  • Figure 1: Dynamic invocation topology in a ByteDance production cluster. (a) Span count distribution per trace, highlighting extreme structural variability. (b) Complementary CDF of call-chain coverage for a key service: 19 dominant patterns account for 90% of all traces.
  • Figure 2: Temporal distribution of dominant call patterns over a 24-hour period, showing strong diurnal periodicity.
  • Figure 3: Workflow of the Morphis framework.
  • Figure 4: Overview of the resource prediction and optimization pipeline.
  • Figure 5: Decomposition of end-to-end latency into self-processing and downstream components. Only self-processing time is used for local resource prediction.
  • ...and 8 more figures