Mar
06

TPS and Nehalem’s in Action

Months ago I thought TPS(Transparent Page Sharing) wasn’t working on our new HP Nehalem blades. To be honest I called VMware Support thinking my View deployment was going to come to a stop. A lot of our VDI blades are sitting on DL460 G1 blades so I never really had the issue before.

Support sorted me out and there has been lots of good articles since; for example:

  • Impact of large pages on consolidation ratios
  •  

  • Large Pages (@gabvirtualworld @frankdenneman @forbesguthrie)  
     
  • VMware KB – Transparent Page Sharing (TPS) in hardware MMU systems
  •  

     

    One thing that I did notice or get annoyed with, is the alarms don’t take into account TPS. vCenter alerts you even though TPS hasn’t been kicked in yet. It would be great to have two separate alerts, one alert without TPS and another when TPS is use and the memory is still high. It would be hard to do because of a sudden spike in usage but I think it would be valuable.

    During non-peak hours I decided to overload a couple of hosts to see how things would play out.

    Before putting two hosts into Maintenance Mode

    Before putting two hosts into Maintenance Mode. Even at around 85% utilization TPS hadn't kicked in.

    Host Using Shared Memory, Two hosts in the cluster now have 60 VM's while the other two sit in maintenace mode.

    TPS Kicking and working it's magic. You can tell that swapping occured while TPS was finding the shared memory.

    I thought it was awesome that with only two hosts, my memory foot print actually went down. This is where you can really save some money on your VDI deployments. The other thing I noticed was that even when the other two hosts where taken out of maintenace mode is that TPS remanied working. I guess they would be one way to fix the alerts!

    TPS Rocks!!!

    Speak Your Mind

    *