Service-Oriented Storage Tiering: A 2014 Approach to a 2004 Problem
Deploying technologies to ensure service levels across the storage infrastructure often encounters difficult practical concerns that must balance the needs of multiple stakeholders in the IT organization. Technology architects need a set of best practices for how these technologies can be deployed across a very wide set of workloads in a standardized, automatable fashion while maintaining the cost and flexibility benefits that come with large-scale storage consolidation.
In this Knowledge Sharing article, awarded 3rd place in the 2014 Knowledge Sharing Competition, Daniel Stafford explores best practices for combining VMAX Host I/O Limits, VMware Storage QoS technologies, SRM reporting, and automation/orchestration technologies to build a Storage Service Catalog that provides consistent performance for multiple tiers for applications living in a shared infrastructure.
Daniel considers an approach to building a multi-tiered storage service catalog which:
• Provides tiers with different costs and performance levels which allow IT to be competitive both internally and externally
• Enforces limits to ensure fair utilization of paid-for resources and reduce the impact of “noisy neighbors”
• Allows for automation to ensure speed and consistency in provisioning
• Allows for flexibility and speed in re-tiering workloads
In the process, Daniel addresses many of the practical problems that come with such a system. These include:
• Automating the analysis new and existing workload to recommend the most effective tier
• Using an SRM solution to proactively detect changing workloads to better partner with application owners
• Providing a consistent, integrated tiering strategy across both virtual and physical hosts
• Algorithms for balancing diverse workloads across available resources at cloud-scale – aka, the “Tetris Problem”
• Array and SAN best practices to avoid limits, ensure flexibility, and maximize uptime
• Array best practices to ensure consistent performance over time – aka, the “First Day is the Best Day Problem”
This article will be of particular interest to architects and IT managers struggling to build a private cloud that supports a wide range of workloads without creating a sprawl of different solutions or increasing costs to the business.