Just Another IT Blog

It's time to share some of my experiences, crazy ideas, tips and tricks !!!

Post Page Advertisement [Top]

Today I was watching the presentation which Jonathan Klick did for VMWorld 2011, “Correctly Sizing Memory in Virtual Environments”

While it brings some good information, metrics and insights about what you should be watching when right sizing your environment, it just mention a lot of terms that you should me familiar with, like : EPT, Large Pages, TPS, etc..

So I decided it was time to shake off the dust and give you some more information about them.

First things first, when monitoring the memory metrics of your guests, understand what each one means: this will help you a lot.

Second, you need to understand what Transparent Page Sharing (TPS) is.
Let’s forget all the bits and bytes behind how a virtual memory page translates into a physical host memory, the important piece to know is that ESX was used to use small pages , 4k in size, so it was easy to ESX to identify common pages and instead of load 2 or more pages with the same data into memory, they just load on physical memory once and share them with the others guests. In basic terms it’s TPS. Easy ?

Now with the advent of new processors like Nehalem some new virtualization functions have been introduced to the processor, one of them is EPT (Extended Page Tables). When ESX identifies that capability on it’s processor they start using Large pages, 2mb in size, instead of using small pages

What Large pages has to do with TPS ? Well, because it’s unlikely that there will be 2 or more 2mb pages identicals, there’s no gain on wasting resources to identify them.
With consequence you will see a lot more host memory consumed and less memory being shared among guests.
So, is there any performance improvement using Large Pages ?
Definitely. Check the VMWare performance evaluation.

But what about TPS, has it been disabled ?
No. It still there, but it will be used more when there’s memory constraints.
When your host is suffering memory pressure the ESX will break these Large pages into small pages, that’s when TPS is back to game finding more easily equal pages and sharing them between guests, increasing the available memory on the host.

Making short a long history.
Keep Large Pages and EPT on. Let ESX take care of the memory management.

As you might know, there’s no size that fits all.
What I want to say with that is, I’m giving you some information about what’s out there, now it’s up to you to understand your environment, your applications, monitor them and tweak them to get the best of your software can offer.

good luck.

More reading:
Large Pages from Yellow Bricks

Monitoring TPS from Yellow Bricks

Large Pages a problem of perception


EPT affects TPS from VMWARE

Bottom Ad [Post Page]