Contact Us

support@onapp.com

U.S: (+1) 888-876-8666

UK: +44 (0) 203-318-5364

Load Balancing new VMs per HV

Completed

Comments

9 comments

  • Avatar
    Stuart Haresnape

    Hi

    This is the same as if you create it through the CP. We're just starting development on a large re-working of some key processes in and around this area and the ability to specify how you load the HVs is part of this so not long now.

    Cheers, Stuart

    0
    Comment actions Permalink
  • Avatar
    Ross Cheetham

    Yep a +1 giant vote for this from myself. This is critical to balancing evenly across our infrastructure. Is this planned for 3.1?

    0
    Comment actions Permalink
  • Avatar
    Ross Cheetham

    Further in regards to the "loading of the HV's", it would be great if you could stipulate a % fill rate, to account for say a failed blade needing to start VM's up across the remaining space available across the remaining infrastructure.

    0
    Comment actions Permalink
  • Avatar
    Ryan MacDonald

    I would say include in this similar logic options for how data stores are filled as they too suffer the same 'load it till its full' setup which is not terribly useful if you want to distribute stuff around.

    The new logic should most importantly also apply to failover situations for hypervisors, from one of my forum posts:
    In OnApp 2.3.x, the failover behavior is such that when a HVM fails, onapp picks a HVM to cold migrate all instances onto and then starts them up. This generally has resulted in more problems than it solves and we in fact completely disable the failover feature on our production HVM's as a result. Basically, there are many occasions where OnApp simply migrates everything to a HVM with a low amount of memory available and/or not enough memory to start all instances back up. Further, migrating 20-30 or more instances all to a single HVM then starting them all on there is often a very bad idea, especially in a failure situation where many of the VM's may be trying to do fsck's and as such causing allot of contention on the storage link for said HVM (yes, even with 10G storage network).

     

    The ability to explicitly assign a single HVM as a "failover" hvm would also be welcomed.

    0
    Comment actions Permalink
  • Avatar
    Daniel Gullin

    +1

    0
    Comment actions Permalink
  • Avatar
    Reliam Engineering

    Thanks!  This is one of our big issues. 

    0
    Comment actions Permalink
  • Avatar
    Stuart Haresnape

    This is in 3.1 to be released very shortly!

    0
    Comment actions Permalink
  • Avatar
    ron

    was this added in 3.1?

    0
    Comment actions Permalink
  • Avatar
    Stuart Haresnape

    Yes - we call it Placement Type.

    https://docs.onapp.com/display/31AG/Create+Hypervisor+Zone

    0
    Comment actions Permalink

Please sign in to leave a comment.