Core Concepts
TierScape proposes a novel solution with multiple software-defined compressed memory tiers that dynamically manages placement and migration of data across compressed tiers to strike the best balance between memory TCO savings and application performance.
Abstract
The paper proposes TierScape, a novel solution with multiple software-defined compressed memory tiers, to tame memory total cost of ownership (TCO) in modern data centers.
Key highlights:
Current state-of-the-art 2-Tier solutions have limited memory TCO savings potential as they only compress cold data, missing opportunities to compress warm data.
TierScape defines multiple compressed tiers in software, each with different compression algorithms, memory allocators, and backing media, enabling flexible trade-offs between memory TCO savings and performance impact.
TierScape employs two data placement models - a waterfall model that gradually moves data to higher TCO saving tiers, and an analytical model that optimizes data placement across tiers based on access patterns and TCO constraints.
Evaluation on real-world benchmarks shows that TierScape can increase memory TCO savings by 22%-40% compared to 2-Tier solutions, while maintaining similar or better performance.
Stats
Memory accounts for 33-50% of the total cost of ownership (TCO) in modern data centers.
TierScape increases memory TCO savings by 22%-40% percentage points compared to state-of-the-art 2-Tier solutions.
TierScape maintains performance parity or improves performance by 2%-10% percentage points compared to 2-Tier solutions.
Quotes
"Memory accounts for 33–50% of the total cost of ownership (TCO) in modern data centers."
"TierScape increases memory TCO savings by 22%–40% percentage points while maintaining performance parity or improves performance by 2%–10% percentage points compared to state-of-the-art 2-Tier solutions."