Tiered Storage Extent Migration via Spike Heat Indicators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tiered storage management systems do not consider non-uniform I/O activity distribution among grains when making data migration decisions, leading to suboptimal storage performance.
Innovation Solution
The system records I/O activity of grains in extents over a predefined time interval, calculates mean heat values and spike heat indicators to identify non-uniform access patterns, and migrates extents based on these metrics to optimize tier placement between SSD and HDD tiers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data migration decisions are based only on mean heat values, then storage tiering is simple to implement, but storage system performance is suboptimal due to non-uniform I/O activity distribution
Solution Approach 1:
The patent segments the I/O activity analysis from the extent level down to the grain level within extents. By calculating mean heat values for each extent and spike heat indicators for each grain, the system creates a hierarchical segmentation that enables more precise migration decisions without overwhelming complexity. This segmentation allows the system to identify non-uniform I/O patterns within extents that would be invisible at the extent level alone.
Solution Approach 2:
The patent applies local quality by treating different grains within the same extent differently based on their individual I/O activity patterns. Instead of migrating an entire extent based on its average heat value, the system identifies specific grains with high spike heat indicators and prioritizes migration of extents containing these hot grains. This local differentiation optimizes storage performance by placing frequently accessed data on faster media while keeping less active data on cheaper media.
2Reliability
If all grains in an extent are treated uniformly, then migration decisions are simple to make, but high I/O workload grains may overload SSDs
Solution Approach 1:
The patent implements local quality by calculating spike heat indicators for individual grains within extents rather than treating all grains uniformly. This allows the system to identify grains with concentrated I/O workloads and prioritize migration of extents containing these hot grains to SSDs. The spike heat indicator specifically measures non-uniform I/O distribution, enabling the system to protect SSDs from overload by proactively migrating extents with high-activity grains before they can overwhelm the faster storage media.
Solution Approach 2:
The patent applies preliminary action by calculating and monitoring spike heat indicators continuously to identify grains that are likely to cause SSD overload. By detecting high spike heat values in advance, the system can proactively migrate affected extents to SSDs before the I/O workload becomes problematic. This predictive approach prevents SSD overload rather than reacting to it after it occurs.
3Productivity
If non-uniform I/O patterns are not detected, then storage tiering is easy to manage, but storage performance is not optimized
Solution Approach 1:
The patent detects non-uniform I/O patterns by calculating spike heat indicators at the grain level within extents. This local analysis reveals I/O concentration patterns that are invisible when viewing only extent-level mean heat values. The spike heat indicator specifically measures the degree of non-uniformity in I/O distribution across grains, enabling the system to identify extents with hot grains that require migration to SSDs for optimal performance.
Solution Approach 2:
The patent adds another dimension to the storage tiering analysis by moving from two-dimensional extent-level heat mapping to three-dimensional grain-level heat mapping within extents. This dimensional expansion allows the system to detect non-uniform I/O patterns that exist within extents, providing a more granular view of data access patterns. The spike heat indicator operates in this additional dimension to identify grains with concentrated I/O workloads.
Data Source
AI summary
Migrating data in a tiered storage system. The tiered storage system includes an upper tier with at least one storage device and a lower tier with at least one storage device. Data on a storage device includes one or more extents, and each extent contains one or more grains. I/O activity of grains is recorded during a predefined time interval. For each extent, a mean heat value over grains in the extent during the predefined time interval is calculated from the recorded I/O activity, where a heat value is a numerical indication of frequency of access of a grain, and a spike heat indicator, which measures non-uniformity of heat values of grains in the extent is calculated. Two extents with mean heat values that differ at most by a predefined threshold are identified, and the extent with a greater spike heat indicator is migrated to an upper tier.