Hardware Feedback Tables for Virtual Machine Core Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In virtual environments, the lack of hardware feedback information to virtual machines leads to inefficient scheduling of threads on hybrid processor cores, resulting in suboptimal performance and increased complexity due to incorrect mapping of virtual processors to physical cores.
Innovation Solution
Providing hardware feedback information directly to virtual machines through HGS feedback tables, allowing them to make informed scheduling decisions based on the dynamic performance and efficiency of physical cores, thereby reducing the need for frequent reshuffling by the root partition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If hardware feedback information is provided only to the root partition (not to virtual machines), then the root partition can schedule threads effectively based on hardware state, but virtual machines cannot make informed scheduling decisions leading to incorrect mapping of virtual processors to physical cores
Solution Approach 1:
The patent introduces HGS feedback tables as an intermediary mechanism that bridges the hardware feedback interface and virtual machines. These tables store hardware feedback information (performance and efficiency metrics) in a format accessible to both the root partition and virtual machines, enabling informed scheduling decisions at both levels without increasing overall system complexity
Solution Approach 2:
The patent segments the hardware feedback information access by creating separate HGS feedback table instances for the root partition and each virtual machine. This segmentation allows each entity to access relevant hardware state information independently, enabling virtual machines to make autonomous scheduling decisions while the root partition maintains overall system coordination
2Ease of operation
If virtual machines schedule threads without hardware feedback information, then the scheduling process is simpler, but the mapping of virtual processors to physical cores becomes incorrect leading to suboptimal performance
Solution Approach 1:
The patent implements a feedback mechanism where hardware feedback information (performance and efficiency metrics from monitor circuits) is stored in HGS feedback tables and made accessible to virtual machines. This feedback enables virtual machine schedulers to make informed decisions about thread-to-core mapping based on actual hardware state, improving both ease of operation and system performance
3Productivity
If the root partition frequently reshuffles virtual processor to physical core mappings to optimize performance, then system performance improves, but the complexity and overhead of management increases
Solution Approach 1:
The patent enables virtual machines to perform preliminary scheduling actions by providing them with hardware feedback information through HGS feedback tables. Virtual machines can pre-optimize their thread-to-core mappings based on hardware state before the root partition needs to intervene, reducing the frequency and complexity of reshuffling operations required by the root partition
Data Source
AI summary
In one embodiment, a method includes: receiving, in a first circuit of a processor, an indication of an entry of the processor into a first virtual machine, the processor to execute in a virtual machine environment having a root partition and at least one virtual partition comprising the first virtual machine; allocating a first hardware feedback structure for the root partition and allocating a second hardware feedback structure for the at least one virtual partition; populating the first hardware feedback structure with first performance and efficiency information regarding a plurality of cores of the processor; populating the second hardware feedback structure with the first performance and efficiency information regarding the plurality of cores of the processor; and providing the first performance and efficiency information from the second hardware feedback structure to a virtual machine scheduler of the at least one virtual partition. Other embodiments are described and claimed.