Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attention Enhanced Entity Recommendation for Intelligent Monitoring in Cloud Systems

Fiza Hussain, Anson Bastos, Anjaly Parayil, Ayush Choure, Chetan Bansal, Rujia Wang, Saravan Rajmohan

TL;DR

DiRecGNN frames the problem of selecting relevant monitoring dimensions in cloud services as a ranking task on a heterogeneous monitor-entity graph. It combines transformer-style graph convolutions, multi-head attention, and random-walk path attention to capture both local and long-range dependencies, with an attention-alignment loss to promote sparse, interpretable attention. The method yields significant gains over baselines in HR, NDCG, and Recall, and is validated with a production user study that reports strong usefulness and calls for end-to-end automation. The work also demonstrates production-scale scalability and practical considerations for deploying dimension recommendations in cloud monitoring workflows.

Abstract

In this paper, we present DiRecGNN, an attention-enhanced entity recommendation framework for monitoring cloud services at Microsoft. We provide insights on the usefulness of this feature as perceived by the cloud service owners and lessons learned from deployment. Specifically, we introduce the problem of recommending the optimal subset of attributes (dimensions) that should be tracked by an automated watchdog (monitor) for cloud services. To begin, we construct the monitor heterogeneous graph at production-scale. The interaction dynamics of these entities are often characterized by limited structural and engagement information, resulting in inferior performance of state-of-the-art approaches. Moreover, traditional methods fail to capture the dependencies between entities spanning a long range due to their homophilic nature. Therefore, we propose an attention-enhanced entity ranking model inspired by transformer architectures. Our model utilizes a multi-head attention mechanism to focus on heterogeneous neighbors and their attributes, and further attends to paths sampled using random walks to capture long-range dependencies. We also employ multi-faceted loss functions to optimize for relevant recommendations while respecting the inherent sparsity of the data. Empirical evaluations demonstrate significant improvements over existing methods, with our model achieving a 43.1% increase in MRR. Furthermore, product teams who consumed these features perceive the feature as useful and rated it 4.5 out of 5.

Attention Enhanced Entity Recommendation for Intelligent Monitoring in Cloud Systems

TL;DR

DiRecGNN frames the problem of selecting relevant monitoring dimensions in cloud services as a ranking task on a heterogeneous monitor-entity graph. It combines transformer-style graph convolutions, multi-head attention, and random-walk path attention to capture both local and long-range dependencies, with an attention-alignment loss to promote sparse, interpretable attention. The method yields significant gains over baselines in HR, NDCG, and Recall, and is validated with a production user study that reports strong usefulness and calls for end-to-end automation. The work also demonstrates production-scale scalability and practical considerations for deploying dimension recommendations in cloud monitoring workflows.

Abstract

In this paper, we present DiRecGNN, an attention-enhanced entity recommendation framework for monitoring cloud services at Microsoft. We provide insights on the usefulness of this feature as perceived by the cloud service owners and lessons learned from deployment. Specifically, we introduce the problem of recommending the optimal subset of attributes (dimensions) that should be tracked by an automated watchdog (monitor) for cloud services. To begin, we construct the monitor heterogeneous graph at production-scale. The interaction dynamics of these entities are often characterized by limited structural and engagement information, resulting in inferior performance of state-of-the-art approaches. Moreover, traditional methods fail to capture the dependencies between entities spanning a long range due to their homophilic nature. Therefore, we propose an attention-enhanced entity ranking model inspired by transformer architectures. Our model utilizes a multi-head attention mechanism to focus on heterogeneous neighbors and their attributes, and further attends to paths sampled using random walks to capture long-range dependencies. We also employ multi-faceted loss functions to optimize for relevant recommendations while respecting the inherent sparsity of the data. Empirical evaluations demonstrate significant improvements over existing methods, with our model achieving a 43.1% increase in MRR. Furthermore, product teams who consumed these features perceive the feature as useful and rated it 4.5 out of 5.
Paper Structure (18 sections, 7 equations, 9 figures, 4 tables)

This paper contains 18 sections, 7 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Formulation
  4. Definition of heterogeneous graph and notations
  5. Data Description and Insights
  6. Attention Enhanced Entity Ranking for Sparse Cloud Graphs
  7. Message Passing Mechanism
  8. Incorporating Random Walk Paths to Capture Global Properties
  9. Training Objective
  10. Experiments
  11. Experiment Setup
  12. Datasets
  13. Training Details
  14. Results (RQ1)
  15. Ablation Study
  16. ...and 3 more sections

Figures (9)

  • Figure 1: a) Monitor entity ranking problem: Nodes represent the monitors, metrics, and dimensions in a cloud setting. Each node contains text features and interacts with their neighboring nodes using link communication and b) The overall service health where the cloud service emits metrics along many dimensions and only few are being used for monitoring health. For eg. in this case the service health depends on health of each datacenter using the service which in turn depends on the health of the individual VMs in the datacenter.
  • Figure 2: Characteristics of the Monitor Entity Graph: a) Distribution of dimension degree based on the metric-to-dimension interactions from the monitor entity graph, and b) Distribution of the percentage of dimensions selected from the set of all dimensions along which the metric is emitted.
  • Figure 3: a) Variation in jaccard similarity of set of dimensions associated with monitors with similar metric, monitor names, and same service account, and b) Distribution of pairwise correlation between dimensions.
  • Figure 4: The overall architecture of DiRecGNN framework. The framework uses an enhanced transformer-style graph convolution over the input graph (block a) which is sampled (block d). Our method incorporates multi-head attention mechanisms (block c), enabling the model to capture the graph structure. Note position encodings (PE) are added for sequence attention but not in case of sparse graph attention. Additionally, we sample random walks (block b) from the target node in order to better capture complex long-range dependencies. The custom losses (block e) include attention-alignment, ranking loss and BCE loss.
  • Figure 5: a) Impact of Ranking Loss on model performance, and b) Impact of Attention-alignment Loss on model performance. Losses improve overall ranking.
  • ...and 4 more figures

Theorems & Definitions (2)

  • Definition 1
  • Definition 2