Browse Prior Art Database

Utilizing AMD (Active Memory De-duplication) to optimize network bandwidth and improve the performance during LPM (Live Partition Migration)

IP.com Disclosure Number: IPCOM000240296D
Publication Date: 2015-Jan-21
Document File: 9 page(s) / 192K

Publishing Venue

The IP.com Prior Art Database

Abstract

The following paper discusses how to utilize AMD (Advanced Memory De-duplication) and prevent de-dupable pages from being copied into the VIOS (Virtual I/O Server) memory area and also prevent such pages from being sent over the network during LPM (Live Pratition Migration).

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 32% of the total text.

Page 01 of 9

Utilizing AMD (Active Memory De-duplication) to optimize network bandwidth and improve the performance during LPM (Live Partition Migration)

Disclosed is a method for improving the performance during LPM (Live Partition Migration) by utilizing AMD (Advanced Memory De-duplication).

Currently during an LPM ( Live partition migration) all the pages from the source VM (virtual machine) are copied to the source VIOS memory and later, all these pages are transferred over the network to the destination hypervisor. This paper address a unique opportunity for optimization and increasing the performance of the LPM operation when AMD is enabled at the destination system.

The existing drawback without AMD in place is, even though the destination server might have similar pages to that of the VM being transferred, all the memory pages of the VM still get copied to the VIOS memory area and are packeted and sent over the network. There is scope for optimization and improving the performance of the LPM which is being addressed in this paper.

Advantages of this proposition are:
1. Improve LPM performance by preventing copy of de-dupe pages to VIOS memory area by the hypervisor.

2. Improve LPM performance by preventing sending of de-dupe pages over the network
3. Lesser VIOS memory space will be utilized.

Abbreviations:

L1-5 in figure 1 = LPARs 1 to 5 {LPAR1 LPAR2 ... LPAR5} at source


D1-7 in figure 1 = LPARs 1 to 7 {LPAR1 LPAR2 ... LPAR7} at destination
S in figure 1 = Set of checksums from the source VIOS for the VM being migrated
D in figure 1= Set of checksums at the destination VIOS of all the VMs/ LPARs at the destination.

Ack = Acknowledgement
LA = Logical address at source / target
RA = Real address at source / target
LAT = Logical address at target
RAT = Logical address at target
C = Checksum at source / target
CT = Checksum at target
AMD = Active Memory De-haduplication
VIOS= Virtual I/O Server
VM=Virtual Machine
LPM= Live partition migration

Note: This idea is applicable only when AMD is enabled at the destination and source. At the source, even if AMD is not enabled completely, then this proposition will work even if the

1


Page 02 of 9

checksum creation module is enabled alone.

Figure 1

In Figure 1, the overall proposal has been captured at a very high level. The proposal is to

prevent the transmission of pages from the source to the destination, that can be de-duped during a Live partition migration. The numbering in the figure show the sequential work flow of the idea. L2 is the VM partition that needs to be migrated from Source VIOS to Destination VIOS. Figure 1 is explained below:

1) Send L2's checksum map: Send the source VM's checksum set which contains the checksum of all its pages to the destination.

2) Send {D-S} or {S ∩ D} or {S-D} list of the checksum from the destination to source vios. The destination sends one of {D-S} or {S ∩ D} or {S-D} list back to the source VIOS, based on which of these has the least elements. It also indicate...