Browse Prior Art Database

Method and System for Managing Paging for Deduplicated Memory

IP.com Disclosure Number: IPCOM000249597D
Publication Date: 2017-Mar-07
Document File: 3 page(s) / 42K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and system is disclosed for managing paging for deduplicated memory. A host system or hypervisor records accesses to deduplicated pages among all virtual machines (VMs) that share pages and applies a paging algorithm, such as least recently used (LRU) or least frequently used (LFU), on a collective basis.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

1

Method and System for Managing Paging for Deduplicated Memory

Some virtual machine (VM) managers support deduplication of VM pages across guests. However, deduplication by VM managers can lead to certain challenges in certain settings.

There exist a number of solutions for paging VM guest memory to disk storage. However, once a page has been deduplicated, a challenge arises as to identifying whether or not it is ready to be stored to disk.

In another existing solution, paravirtualization allows for some component of a guest to be aware of some aspects of an underlying host. However, prior art solutions that consider virtual resource management in light of host-guest awareness turn to criteria such as Quality of Service (Qos) considerations for provisioning memory at a granularity of a page at a time.

Though there are existing solutions for resource management and for deduplication, very few solutions involve paravirtualization applied to virtual memory paging.

Disclosed is a method and system for managing paging for deduplicated memory. A host system or hypervisor records accesses to deduplicated pages among all VMs that share pages and applies a paging algorithm, such as least recently used (LRU) or least frequently used (LFU), on a collective basis.

In accordance with the method and system, the host system or hypervisor can track deduplicated pages by tracking a status for each guest OS referencing the page, as to whether that guest has directed the page to be paged out, or at least whether the page meets its page-out criteria, and also can track a set of file references for any page for which all guests agree that criteria are met. A primary file reference can reflect an actual backing store for the page. Additional file references can reflect other (potentially guest-specific) backing stores that can become primary in case the guest having the primary reference accesses any page for which at least two other guests each have a reference, or the process having the primary reference is terminated, or the guest having the primary reference is shut down.

Paging of deduplicated memory for processes running under guest operating systems acting across a distributed elastic cloud system can include a paravirtual memory manager, running on each guest, that enables a host system or hypervisor to compare memory content across processes and guests in light of paging algorithms that can vary from guest to guest, The memory manager can include a means for capturing a guest’s directive or criteria to page out certain memory, and a deduplication system that tracks guest readiness to page out deduplicated memory, that pages out the

2

deduplicated memory when all guests agree to do so, and that tracks one or more file references for each deduplicated page.

Fig. 1, illustrates the method and system for managing paging for deduplicated memory.

Figu...