Browse Prior Art Database

System and Method for Memory Performance and Power Management in Cloud Infrastructure

IP.com Disclosure Number: IPCOM000239585D
Publication Date: 2014-Nov-17
Document File: 3 page(s) / 46K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is an idea in the area of server memory subsystem that provides consolidation/expansion of memory usage to achieve power savings/performance goals

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 3

System and Method for Memory Performance and Power Management in Cloud Infrastructure

In server memory subsystem, interleaving read and write operations across multiple channels provide significant performance benefits. On the other hand, interleaving needs several memory components and channels to be operational and, therefore, has more power consumption.

Non-interleaving scheme allow operations with less number of channels, which would help save power; however, it has adverse effects on performance as data have to be sequentially accessed.

    Traditionally, during boot of a computer, memory channels are configured to have either interleaving or non-interleaving schemes, depending on the performance requirements. In cloud computing infrastructures, it is very hard to predict the memory utilization pattern as the Virtual Machines (VM) may have workloads with different utilization requirements. However, Service Level Agreement (SLA) would have clear specifications indicating performance requirements and utilization patterns. VM Provisioning Manager (VPM) controls the hardware layer attributes and set ups the right configuration to meet the SLA targets. Based on dynamic provisioning requirements, VPM has the requirement of hosting VM pools to a specific memory hardware. A VM pool may have the SLA target of either performance improvement or power conservation.

    VPM needs to have the following features on the memory subsystem to enable seamless "dynamic provisioning." Facility to migrate data seamlessly to carry out the data consolidation process. VPM has the need to interchange between power and performance modes, at run time. Server systems typically are set to operate with more memory channels (interleaving enabled) to deliver maximum performance. Server systems have varying patterns of memory usage ranging from low to high utilization. Low utilization scenarios provide opportunities to consolidate usage from "many" to "few" memory modules and turn the unused modules/channels to low-power (LP) mode. However, this consolidation process has the following major challenges. Data transfer carried out by software stack at high levels has huge time penalty, and performance would severely be affected during this process. Once data is consolidated, changing from "interleaving to non-interleaving" (or vice versa) requires reboot of the system.

    Disclosed is an idea in the area of server memory subsystems which aims to enable a feature that provides consolidation/expansion of memory usage to achieve power savings/performance goals.

A data consolidation controller (DCC) as part of memory buffer. It further comprises two sub units, data migration unit (DMU) and channel setup unit (CSU) to support data co...