LLC Insertion Bypassing for Probe Filter Stress Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-node computer systems, probe filters can face capacity issues due to heavy index-conflicts and sparse accesses, leading to premature eviction of cache lines from the last level cache (LLC), which degrades system performance.
Innovation Solution
Implementing a system probe aware last level cache insertion bypassing policy, where the cache is partitioned into two portions based on the hit rate and recall probe rate. If the hit rate of the first portion is high, a non-bypass insertion policy is applied to the second portion, while a bypass insertion policy is used if the hit rate is low, reducing recall probes and stress on the probe filter.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a probe filter is sized to cover all caches for expected traffic patterns, then memory bandwidth is improved, but capacity issues occur with nonstandard traffic patterns causing premature eviction
Solution Approach 1:
The probe filter capacity is made dynamic by allowing it to expand when recall traffic is detected. The system monitors for recall traffic patterns and increases probe filter capacity when needed, rather than using a static size. This resolves the contradiction by providing sufficient capacity for nonstandard traffic while maintaining efficient bandwidth for normal traffic patterns.
Solution Approach 2:
The system changes the parameter of probe filter capacity based on traffic conditions. When recall traffic is detected, the probe filter capacity is increased to prevent premature eviction. This parameter change allows the system to adapt to different traffic patterns and maintain both memory bandwidth efficiency and cache line retention reliability.
2Reliability
If recall traffic is handled with increased probe filter capacity, then cache line eviction is reduced, but system complexity increases
Solution Approach 1:
The system uses self-service by having the probe filter monitor its own recall traffic conditions and automatically adjust its capacity when needed. The probe filter detects recall traffic patterns and triggers capacity expansion without requiring complex external control mechanisms, thereby maintaining reliability while limiting the increase in system complexity.
Solution Approach 2:
The system implements feedback by monitoring recall traffic conditions and using this information to adjust probe filter capacity. The probe filter detects when recall traffic is occurring and triggers an increase in capacity to prevent premature eviction. This feedback mechanism provides a simple way to maintain cache line retention without requiring complex proactive management.
3Device complexity
If LLC lines are prematurely evicted to make room in probe filter, then probe filter capacity is maintained, but system performance degrades
Solution Approach 1:
The system applies preliminary anti-action by preventing premature eviction through proactive detection of recall traffic conditions. When recall traffic is detected, the probe filter capacity is increased before eviction can occur, thereby protecting system performance. This preliminary action prevents the harmful effect of premature eviction without requiring complex reactive measures.
Data Source
AI summary
Systems, apparatuses, and methods for employing system probe filter aware last level cache insertion bypassing policies are disclosed. A system includes a plurality of processing nodes, a probe filter, and a shared cache. The probe filter monitors a rate of recall probes that are generated, and if the rate is greater than a first threshold, then the system initiates a cache partitioning and monitoring phase for the shared cache. Accordingly, the cache is partitioned into two portions. If the hit rate of a first portion is greater than a second threshold, then a second portion will have a non-bypass insertion policy since the cache is relatively useful in this scenario. However, if the hit rate of the first portion is less than or equal to the second threshold, then the second portion will have a bypass insertion policy since the cache is less useful in this case.