Nice article on how Linux Kernel Shared Memory (KSM) works

[Ninjix]

While rolling through my RSS feeds this morning, I came across this informative article on how Linux Kernel Shared Memory (KSM) works.

http://www.ibm.com/developerworks/linux/library/l-kernel-shared-memory/index.html

I've been really impressed with KSM and especially the Proxmox-ve-2.6.32 implementation. KVM on a KSM enabled hosts seem much more responsive than VMware ESX on the same hardware running over-commited. I think ESX pushes more off to disk page for their memory de-dup. The ESX VM always respond sluggishly after having been near-idle for an extended period of time. My Windows KVM on Proxmox respond right away no matter how long they've been idling. This is especially noticeable on Monday mornings when I make my rounds tending different servers.

Keep up the great work Proxmox Team.

 

[blender]

While rolling through my RSS feeds this morning, I came across this informative article on how Linux Kernel Shared Memory (KSM) works.

Cool, thanks for sharing.

In the Resources section there is the statement:

"In the KSM section, you'll find an interesting reference to running 52 Windows® XP VMs on a single server with 16GB of RAM (where each was allocated its own 1GB RAM segment but shared with KSM)."

Wow, that's pretty impressive!

Anyone have Linux density numbers?

Cheers

 

[Ninjix]

I've tested 6 Win2k8 1gb on a 2gb Proxmox-ve-2.6.32 host in our lab. I loaded each with a +-15 minute delay between qm start and then a one hour idle. Performance wasn't that bad considering and good enough to win my executives over. In production things aren't so rosy due to application memory volatility and churn but I can see the memory de-dup of the "static" Windows OS memory.