Workload Placement via Access Locality Sampling

Solution Overview

Problem

Existing computing systems with disaggregated storage arrangements face challenges in adapting to changing workloads, leading to resource overuse and performance issues due to initial configuration inadequacies and reliance on lagging performance indicators like I/O operations and latency.

Innovation Solution

The use of access locality measures, such as reuse distance, and statistical techniques to detect changes in workload patterns in real-time, enabling dynamic workload placement and resource optimization across compute nodes, and the development of performance models to predict resource needs based on access locality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If initial configuration is used for disaggregated storage arrangements, then system setup is simplified, but the system cannot adapt to changing workloads leading to resource overuse and performance issues

Engineering Contradiction:

Improveadaptability to changing workloadsVSAvoidsystem configuration complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs self-characterization by automatically measuring access locality metrics and workload patterns without external intervention. The storage system autonomously generates performance models and detects workload changes, enabling adaptive resource allocation without complex manual configuration or external control systems.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system pre-characterizes workload access patterns by measuring access locality during initial operation. Performance models are generated in advance based on observed workload behavior, allowing the system to proactively adapt to changing workloads before performance degradation occurs, rather than reacting to lagging indicators.

Inventive Principle:

Principle #10Preliminary action

2Speed

If lagging performance indicators like I/O operations and latency are used, then performance measurement is straightforward, but the system cannot rapidly adapt to changing workloads

Engineering Contradiction:

Improvespeed of workload adaptationVSAvoidworkload characterization precision

Core Design Contradiction:

SpeedVSMeasurement precision

Solution Approach 1:

The system pre-characterizes workload access patterns by measuring access locality during initial operation. Performance models are generated in advance based on observed workload behavior, allowing the system to proactively adapt to changing workloads before performance degradation occurs, rather than reacting to lagging indicators.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Instead of measuring traditional lagging performance indicators like I/O operations and latency, the system inverts the approach by measuring access locality metrics (reuse distance, access patterns) that lead performance changes. This allows prediction of workload changes before they manifest as performance degradation.

Inventive Principle:

Principle #13The other way round (Inversion)

3Productivity

If resource capacities are fixed, then system management is simpler, but resource utilization is suboptimal under varying workload conditions

Engineering Contradiction:

Improveresource utilization efficiencyVSAvoidresource management complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system performs self-characterization by automatically measuring access locality metrics and workload patterns without external intervention. The storage system autonomously generates performance models and detects workload changes, enabling adaptive resource allocation without complex manual configuration or external control systems.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system dynamically adjusts resource allocation based on real-time workload characterization. Compute nodes are selected for workload placement based on current access locality patterns and performance models, allowing resource capacities to adapt flexibly to varying workload conditions while maintaining optimal utilization.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS20230325257A1Workload measures based on access locality

Publication Date: 2023.10.12 HEWLETT PACKARD ENTERPRISE DEV LP

AI summary

In some examples, a system samples a subset of input/output (I/O) accesses of a storage, the I/O accesses being part of a workload. The system determines, based on the sampled subset of the I/O accesses, a first reuse distance distribution for a first time interval, determines a similarity measure representing a similarity of the first reuse distance distribution and a second reuse distance distribution for a second time interval different from the first time interval, and based on a change in the similarity measure, triggers a workload placement process to determine a placement of the workload on a compute node of a plurality of compute nodes.