Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dodoor: Efficient Randomized Decentralized Scheduling with Load Caching for Heterogeneous Tasks and Clusters

Wei Da, Evangelia Kalyvianaki

TL;DR

Dodoor tackles efficient task scheduling in heterogeneous data centers by replacing per-request probing with batched, centrally pushed load caches. It introduces a novel anti-affinity resource-load score (RL) and uses a two-choice selection to place tasks, balancing resource demands and execution durations. Empirical results on a 101-node cluster show up to 55–66% reduction in scheduling messages, up to 33.2% throughput gains, and notable improvements in mean and tail makespan across Azure VM traces and serverless FunctionBench workloads. The design emphasizes resilience and scalability through asynchronous updates and a push-based data store, offering practical benefits for real-world cloud environments.

Abstract

This paper introduces Dodoor, an efficient randomized decentralized scheduler designed for task scheduling in modern data centers. Dodoor leverages advanced research on the weighted balls-into-bins model with b-batched setting. Unlike other decentralized schedulers that rely on real-time probing of remote servers, Dodoor makes scheduling decisions based on cached server information, which is updated in batches, to reduce communication overheads. To schedule tasks with dynamic, multidimensional resource requirements in heterogeneous cluster, Dodoor uses a novel load score to measure servers' loads for each scheduled task. This score captures the anti-affinity between servers and tasks in contrast to the commonly used heuristic of counting pending tasks to balance load. On a 101-node heterogeneous cluster, Dodoor is evaluated using two workloads: (i) simulated Azure virtual machines placements and (ii) real serverless Python functions executions in Docker. The evaluation shows that Dodoor reduces scheduling messages by 55--66% on both workloads. Dodoor can also increase throughput by up to 33.2% and 21.5%, reduce mean makespan latency by 12.1% and 7.2%, and improve tail latency by 21.9% and 24.6% across the two workloads.

Dodoor: Efficient Randomized Decentralized Scheduling with Load Caching for Heterogeneous Tasks and Clusters

TL;DR

Dodoor tackles efficient task scheduling in heterogeneous data centers by replacing per-request probing with batched, centrally pushed load caches. It introduces a novel anti-affinity resource-load score (RL) and uses a two-choice selection to place tasks, balancing resource demands and execution durations. Empirical results on a 101-node cluster show up to 55–66% reduction in scheduling messages, up to 33.2% throughput gains, and notable improvements in mean and tail makespan across Azure VM traces and serverless FunctionBench workloads. The design emphasizes resilience and scalability through asynchronous updates and a push-based data store, offering practical benefits for real-world cloud environments.

Abstract

This paper introduces Dodoor, an efficient randomized decentralized scheduler designed for task scheduling in modern data centers. Dodoor leverages advanced research on the weighted balls-into-bins model with b-batched setting. Unlike other decentralized schedulers that rely on real-time probing of remote servers, Dodoor makes scheduling decisions based on cached server information, which is updated in batches, to reduce communication overheads. To schedule tasks with dynamic, multidimensional resource requirements in heterogeneous cluster, Dodoor uses a novel load score to measure servers' loads for each scheduled task. This score captures the anti-affinity between servers and tasks in contrast to the commonly used heuristic of counting pending tasks to balance load. On a 101-node heterogeneous cluster, Dodoor is evaluated using two workloads: (i) simulated Azure virtual machines placements and (ii) real serverless Python functions executions in Docker. The evaluation shows that Dodoor reduces scheduling messages by 55--66% on both workloads. Dodoor can also increase throughput by up to 33.2% and 21.5%, reduce mean makespan latency by 12.1% and 7.2%, and improve tail latency by 21.9% and 24.6% across the two workloads.
Paper Structure (26 sections, 2 equations, 8 figures, 4 tables, 1 algorithm)

This paper contains 26 sections, 2 equations, 8 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Balls-into-bins Problem
  4. Power-of-two in Modern Schedulers
  5. Variants of Randomized Scheduler
  6. Scope and Proposal
  7. Dodoor Scheduling
  8. System Model
  9. Scheduling Algorithm
  10. System Design
  11. Scheduler and Data APIs
  12. Data Store
  13. Scheduler
  14. Server Execution
  15. Cluster Operation and Reliability
  16. ...and 11 more sections

Figures (8)

  • Figure 1: Probing strategy
  • Figure 2: Dodoor Framework
  • Figure 3: Azure Exp: VM Lifetime Distribution
  • Figure 4: Azure Exp: Aggregated Scheduling Metrics Under Varying QPS
  • Figure 5: Azure Exp: Average Resource Variance and Utilization Under Varying QPS
  • ...and 3 more figures