[pve-devel] mass deployment for testing ipcc_send_rec failed

Daniel Hunsaker danhunsaker at gmail.com
Wed May 28 22:19:56 CEST 2014


> Disks are on external ceph system - so compute node is not under load. 15
VMs for 48 cores are not much.
>
At that point you're waiting for the network I/O, so there's still that
(which is sometimes actually a bit slower).  Plus, the VM's devices will
still be loaded from the compute node itself, unless it's a completely
diskless server (which has its own potential problems).  So you still have
some load to consider.  Add that to the amount of other operations
happening at the same time (file locking/unlocking, waiting for files to be
unlocked, virtual hardware initialization, resource allocation, etc) and
starting a VM is computationally expensive, even with 48 cores.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.proxmox.com/pipermail/pve-devel/attachments/20140528/1344e045/attachment.htm>


More information about the pve-devel mailing list