TPS and memory pressure guest testing

[jrr001]

I am looking for a good way to test vSphere 5.1 memory reclamation in guest VMs but need a good tool to apply memory pressure in the guest.

Does anyone have a tool to suggest for this type of testing. At VMworld they would have a memhog type utility in the performance labs.

Appreciate any suggestions or tweaks to optimize the guest memory reclamation process.

[Dave_Mishchenko]

If you're dealing with a single VM you can apply a memory limit to it which will force the balloon driver into play.  This utility can be used to consume memory within a VM - http://www.jam-software.com/heavyload/.   But if you want to see something a bit more real world,  create a single VM with a large memory reservation that'll consume a good part of a hosts memory.  Then start a memory utility in it like memtest86.com.  As you start or run other VMs you'll start to see how ESXi uses compression, ballooning and swaping.

If the below image the host has 20 GB memory and was showing about 17 GB in use and 16  GB available in resource allocation.   I turned on a VM with 16 Gb and started a memory test and immediately ESXi had to start swapping to keep up with a memory requirements.

[nive1103]

Hi,

Have a look athttp://www.yellow-bricks.com/2010/04/09/generating-load/ for simulating load.

Make sure VM tools is installed in guest for better memory reclamation. This will enable memory ballooning and also provide accurate guest analysis data.

[jrr001]

In trying some memory pressure tests, guest OS is server2012 with 2GB RAM, I use the heavload tool to fill memory and leave only 512mb free in the guest VM OS. vSphere 5.1 does not appear to come back and reclaim that RAM after filling it and releasing the RAM.

On got very good results from TPS just running 4 of these guest VMS (2GB each) on a 6GB Nested vSphere 5.1 server (saved about 3,234mb) out of the 6GB.

Am I missing something here on how this should work? Attached is jpg file showing the guest memory usage from VMware perspective.

Thanks.