L3 Cache Partitioning for QoS Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data centers, the imbalance in Quality of Service (QoS) requirements among applications and shared resources in level 3 (L3) caches leads to inefficient processor utilization, as applications with higher QoS requirements cannot be run efficiently on compute units already running lower QoS applications due to resource sharing.
Innovation Solution
The L3 cache is partitioned into subcaches, with each compute unit allocated specific subcaches based on QoS requirements, using a mapping table to manage associations between compute units and subcaches, allowing for efficient scheduling and data storage without interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If applications are run on shared compute units with common L3 cache, then processor utilization increases, but QoS requirements cannot be met due to resource interference
Solution Approach 1:
The L3 cache is divided into multiple partitioned subcaches, each associated with a specific compute unit. This segmentation allows each compute unit to have dedicated cache space, eliminating interference from other applications while maintaining high processor utilization through shared compute units.
Solution Approach 2:
Different partitions of the L3 cache are allocated to different compute units based on their specific QoS requirements. High QoS applications receive dedicated cache partitions that guarantee performance, while lower QoS applications share remaining resources, ensuring each application receives appropriate quality of service.
2Reliability
If L3 cache is partitioned into subcaches for each compute unit, then QoS requirements are met, but cache management complexity increases
Solution Approach 1:
Each compute unit autonomously manages its own associated subcache partition, making independence decisions without requiring complex centralized cache management. The compute unit itself serves its cache needs, reducing overall system complexity while maintaining QoS guarantees.
3Productivity
If compute units are shared across multiple applications, then resource efficiency improves, but interference between applications with different QoS occurs
Solution Approach 1:
The L3 cache is segmented into separate partitions for each compute unit, preventing interference between applications running on the same compute unit. Each application accesses only its dedicated cache partition, eliminating harmful interference while maintaining resource sharing at the compute unit level.
Solution Approach 2:
The partitioned cache structure acts as an intermediary between compute units and main memory, providing isolation between applications. Each compute unit interacts only with its own cache partition, which mediates access to the broader memory system, preventing direct interference between applications.
Data Source
AI summary
A method and apparatus for partitioning a cache includes determining an allocation of a subcache out of a plurality of subcaches within the cache for association with a compute unit out of a plurality of compute units. Data is processed by the compute unit, and the compute unit evicts a line. The evicted line is written to the subcache associated with the compute unit.