Processor Cache Extension via Donor Ownership Transfer

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In multi-processor computers, limited cache capacity leads to increased cache misses and performance loss due to higher latency times in accessing memory, as each processor has a dedicated point-to-point connection to memory modules, resulting in inefficient utilization of available space.

Innovation Solution

Implementing a cache extension mechanism where a donor processor detects non-use of its cache, broadcasts availability, and transfers ownership to a requesting processor through a series of messages, allowing dynamic cache extension and optimization of cache usage across processors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Volume of stationary object

If cache capacity is limited to reduce silicon footprint, then memory module size is reduced, but cache miss rate increases leading to performance loss

Engineering Contradiction:
Improvecache memory sizeVSAvoidprocessor performance
Core Design Contradiction:
Volume of stationary objectVSProductivity

Solution Approach 1:

The patent merges cache resources from multiple processors into a shared pool. When a processor experiences a cache miss, it can access cache lines from other processors' caches through inter-processor communication, effectively combining available cache capacity across the system to reduce overall cache miss rates while maintaining individual processor cache sizes.

Inventive Principle:
Principle #5Merging (Combining)

2Speed

If each processor has dedicated point-to-point connection to memory modules, then memory access is direct, but cache utilization efficiency decreases

Engineering Contradiction:
Improvememory access speedVSAvoidcache resource utilization efficiency
Core Design Contradiction:
SpeedVSLoss of energy

Solution Approach 1:

The patent makes cache memory universal by enabling any processor to access any cache in the system, not just its own dedicated cache. The cache coherence protocol allows cache lines to be shared across multiple processors, making the cache resource multi-functional and accessible to multiple processors dynamically based on demand, thereby improving utilization efficiency while maintaining fast access through existing point-to-point connections.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentUS8924644B2Extending cache in a multi-processor computer
Publication Date: 2014.12.30 LENOVO GLOBAL TECHNOLOGIES SWITZERLAND INTERNATIONAL GMBH
  • US8924644B2 patent drawing
  • US8924644B2 patent drawing
  • US8924644B2 patent drawing

AI summary

Methods, apparatuses, and computer program products of extending cache in a multi-processor computer are provided. Embodiments include detecting, by a donor processor, nonuse of a donor processor's cache; broadcasting to one or more processors in the multi-processor computer, by the donor processor, a donor-ready message indicating the donor processor's cache is available for ownership transferment; receiving from a first requesting processor, by the donor processor, a first ownership-request message requesting ownership of the donor processor's cache by the first requesting processor; transmitting to the first requesting processor, by the donor processor, an ownership-grant message indicating an intention of the donor processor to transfer ownership of the donor processor's cache to the first requesting processor; and receiving from the first requesting processor, by the donor processor, an ownership-claim message indicating that the first requesting processor intends to claim ownership of the donor processor's cache.