Simultaneous Multithreading Cache Segmentation for Resource Aggressive Applications
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern computing systems with multiprocessing and simultaneous multithreading often experience cache thrashing due to resource aggressive applications, leading to reduced performance.
Innovation Solution
A system daemon is used to balance resource aggressive applications across system resources by obtaining resource and process information, redistributing processes based on measurements, and offlineing sibling resources to minimize cache thrashing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple virtual logical processors share caches in a simultaneous multithreading system, then system resource utilization is improved, but cache thrashing occurs leading to reduced performance
Solution Approach 1:
The patent segments the cache memory into separate cache spaces associated with different physical processors. Each physical processor has its own dedicated cache space, preventing other processors from accessing or interfering with it. This segmentation eliminates cache thrashing while allowing multiple processors to operate simultaneously, resolving the contradiction between resource utilization and performance.
Solution Approach 2:
The patent introduces a cache space as an intermediary structure between physical processors and main memory. Each physical processor gets a dedicated cache space that acts as a buffer, isolating cache accesses to specific processors and preventing the cache thrashing phenomenon that would otherwise occur when multiple processors share the same cache.
2Adaptability or versatility
If resource aggressive applications are allowed to execute on multiple logical processors, then parallelization capability is improved, but cache interference increases
Solution Approach 1:
The patent segments cache resources by physically associating dedicated cache spaces with each physical processor. This ensures that even when multiple logical processors execute resource aggressive applications, they cannot interfere with each other's cache spaces, eliminating cache interference while preserving parallelization capability.
Solution Approach 2:
The patent implements local quality by giving each physical processor exclusive access to its own cache space. This localized cache access pattern ensures that cache behavior is optimized for each individual processor's workload, preventing the harmful cache interference that would occur with shared caches.
3Adaptability or versatility
If unscalable processes are allocated to shared resources, then system flexibility is improved, but cache thrashing occurs
Solution Approach 1:
The patent segments cache access rights by physically binding dedicated cache spaces to each physical processor. When unscalable processes are allocated to physical processors, they automatically receive exclusive access to that processor's dedicated cache space, eliminating cache thrashing while maintaining system flexibility in process allocation.
Solution Approach 2:
The dedicated cache space acts as an intermediary that isolates unscalable processes from cache interference. Even though these processes run on shared physical hardware, the dedicated cache space mediates their memory accesses, preventing thrashing and maintaining performance.
Data Source
AI summary
A simultaneous multithreading computing system obtains process information for the simultaneous multithreading computing system. The process information comprises a plurality of processes associated with the simultaneous multithreading computing system. The simultaneous multithreading computing system obtains resource information for the simultaneous multithreading computing system. The resource information comprises a plurality of available resources in the simultaneous multithreading system. The simultaneous multithreading computing system determines that a process from the plurality of processes is unscalable on the simultaneous multithreading computing system. Upon determining that the process is unscalable, the simultaneous multithreading computing system selects a resource to execute the unscalable process based on the resource information. Upon determining that a sibling resource is associated with the selected resource, the simultaneous multithreading computing system disconnects the sibling resource.