Browse Prior Art Database

Memory Thermal Management Policy Using a Mirrored Mode Operation

IP.com Disclosure Number: IPCOM000034008D
Publication Date: 2005-Jan-11
Document File: 3 page(s) / 25K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method that uses the following components for a three-level thermal management policy:

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 45% of the total text.

Memory Thermal Management Policy Using a Mirrored Mode Operation

Disclosed is a method that uses the following components for a three-level thermal management policy:

 

  • A memory controller and memory subsystem that supports a mirrored-mode of operation
  • On-die thermal sensors placed on the MCH die, the DIMM or DRAM devices, and/or the buffer controllers
  • Configuration and firmware methods in the form of a hierarchy of three policy states

Benefits include thermal management without the associated thermal throttling.

Background

Currently, memory thermal management strategically places a thermal sensor at the hot spot of either a component or a system.  The thermal sensor is set by software to some applicable thermal trip point, so that if the sensor detects an over-temperature condition it can set off a series of events to correct the condition. These events involve a technology known as thermal throttling. Thermal throttling “throttles” back the number of memory transactions that are processed per a fixed period of time, based on the current or projected temperature that is detected at the thermal sensor. As the number of memory transactions increases, the temperature measured at either the memory controller die or the DRAM devices increases. Unfortunately, thermal throttling reduces the number of memory transactions allowed in order to reduce the temperature to an acceptable level; this results in a decreased overall performance in the memory subsystem. 

General Description

The disclosed method develops both silicon and firmware technology to create more efficient methods for thermal management, without incurring the performance penalties associated with the current state of the art.

Note. The disclosed method does not include any improvements to traditional mechanical thermal solutions, such as heat sinks, liquid cooling, or airflow.

Memory Read Transaction Redirection, Level One Thermal Management Policy. Overall memory traffic is made up of memory read and memory write transactions. In mirrored mode operation (also known commonly as RAID Level 1), the memory subsystem is divided such that there is a back-up copy of all unique data; if a memory read failure occurs to one copy of the data, the read transaction can still be completed successfully by redirecting the read to the back-up or mirrored copy.  Memory-related thermal conditions are a by-product of both read and write operations taken into account over a period of time. Current memory controllers that support mirrored mode do so by directing all write transactions to both mirrors, thereby creating the original and spare copy.  But read transactions are typically only directed to what is known as the primary mirror. They are only redirected to the secondary mirror in the case of permanent data failures. The combination of both reads and writes to the primary mirror can reach a threshold such that the thermal trip point is reached or exceeded. 

The disclosed method replaces the cur...