Browse Prior Art Database

Method and Apparatus to enhance memory RAS on servers using NVM Express link

IP.com Disclosure Number: IPCOM000247925D
Publication Date: 2016-Oct-11
Document File: 2 page(s) / 337K

Publishing Venue

The IP.com Prior Art Database

Abstract

Abstract: NVMe (Non-volatile Memory Express) is a communication interface similar to SATA or SAS which supports data transfer between main memory & secondary memory at faster speeds with variants of queue lengths and queue size. The proposed solution is to include "smart logic hybrid controller" which would determine scenarios where memory is inappropriately utilized and use the NVMe link to transfer the to/from main memory and secondary memory.

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

Page 01 of 2

Method and Apparatus to enhance memory RAS on servers using NVM Express link

Title: Method and Apparatus to enhance memory RAS on servers using NVM Express link

Introduction:

Technology scaling of DRAM is nearing the end, prompting the need for a replacement technology. Bandwidth requirements in today's systems are huge and DRAMs have been unable to keep up. 3D or stacked DRAM die and various other next-generation memory types are being developed/ explored. Four main technical challenges with the DRAM - power consumption, bandwidth, latency and scaling. There is no replacement technology available to match the speed, density and cost of DRAMs and scaling DRAM to 1znm (12-14nm) is possible, but not beyond. The transition from 20nm (today) to 1znm will need improved process and also, pose integration challenges. NVMe is a communication interface similar to SATA or SAS; supports data transfer between main memory & secondary memory at faster speeds with variants of queue lengths and queue size. This interface uses pipelining and is compatible with industry standard PCIe technology.

Problem Statement:

A server system has main memory in the order of approximately 500GB to 4TB or even higher. At any instance, it is possible that the entire memory is


 Allocated to application(s) but not utilized


 Allocated to application(s) but utilized rarely


o In any of the above cases, the memory is idle but still consuming power - scrub / refresh to keep the DRAM's active


 Allocated to application(s) and utilized heavily, which means,


o It requires 1 additional level of RAS - sparing to prevent CIE


o It might need memory mirroring capabilities but unable to do so; since it was not enabled at IPL / power-on


o It needs more memory but not alloca...