Multi-tenant ETL Workload Management via Dynamic Throttling

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In multi-tenant ETL systems, one tenant's high volume of data can monopolize processing resources, delaying the work of other tenants, leading to unfair service distribution among tenants within the same category.

Innovation Solution

A method and system that determine the total workload and queued workload of a multi-tenant ETL system to delay the addition of work requested by a tenant, ensuring fairness by controlling the delivery of work and preventing overload, using a workload component, analysis component, and delay component to compute and apply a delay coefficient based on the system's load and the tenant's workload.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If one tenant submits large amounts of work to the ETL system, then that tenant's data processing capacity is improved, but other tenants experience delayed processing and unfair service distribution

Engineering Contradiction:

Improvedata processing capacityVSAvoidprocessing delay for other tenants

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The system continuously monitors the total workload in the queue and uses this feedback to dynamically adjust the rate at which new work is accepted from each tenant. When the queue reaches a threshold, the system provides feedback by delaying or throttling incoming work requests, preventing any single tenant from monopolizing processing resources while maintaining high overall productivity

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The workload acceptance rate is made dynamic rather than static. The system adjusts the rate of accepting work from each tenant based on real-time conditions of the processing queue. This dynamic adjustment allows the system to accommodate high-volume tenants when capacity is available while automatically reducing their rate when the system is near capacity, preventing processing delays for other tenants

Inventive Principle:

Principle #15Dynamics

2Productivity

If the system accepts all incoming work requests without restriction, then overall system utilization is maximized, but service fairness among tenants deteriorates

Engineering Contradiction:

Improvesystem utilizationVSAvoidservice fairness

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system changes the parameter of workload acceptance from an open state to a controlled state based on queue depth. By monitoring the total workload parameter and comparing it against thresholds, the system dynamically adjusts the acceptance parameter for each tenant, maintaining high system utilization while ensuring service fairness through parameter-based control

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The workload queue acts as an intermediary between incoming work requests and the processing engine. By introducing this intermediary buffer and monitoring its state, the system can mediate between the desire to accept all work (for high utilization) and the need to maintain fairness (by controlling what gets added to the queue based on current load conditions)

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11042415B2Multi-tenant extract transform load resource sharing

Publication Date: 2021.06.22 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Concepts for sharing processing resource of a multi-tenant extract transform load, ETL, system are presented. In such concepts, a total workload of the multi-tenant ETL system is considered along with the queued workload of tenant in order control deliver of the queued workload to the system. Such control is undertaken, for example, by delaying the work of the tenant. Proposed embodiments therefore seek to devise a policy to achieve fairness amongst tenants.