Hi Dmitry, did you roll all 450 VMs from the same template (450 VMs, on single host ???)
With KVM on local/shared storage, it works by initially moving/copying a template qcow2 image from Secondary Storage NFS, to Primary Storage (local or shared NFS...) and then protecting this image and making it parent image for all VM's images(root disks). When very first VM is deployed from some template - this qcow2 copy process (as described above) happens and then there is new qcow2 image created and linked to the parent image (google on qcow2 backing file for explanation). I assume 450 VMs (child images) being linked to same parent qcow2, can make issue - I vaguely remember reading somewhere that it doesn't scale that well - but I might be wrong. If 450 VMs on single host (I can't imagine that really, unless heavy, heavy CPU overprovisioning...) than this also might have something to do with it. Cheer Andrija On Thu, 5 Jul 2018 at 09:46, Dmitry Berezhnoy <[email protected]> wrote: > Hello, > > I have installed cloudstack-management, cloudstack-agent(KVM hypervisor) > and nfs-server at the same physical server. Then I have created virtual > machines one by one in 8 parallel threads. For each thread I created shared > network(VR). This server has two SSDs with ZFS filesystem. Copying files > and disk speed is good enough (not only in system idle), but when I have > 450 virtual machines the next one has been creating during 5 minutes, is it > correct work? This time becomes too long on 200 (about a minute) virtual > machines and increase on every 10 machines or so. Does anybody have any > ideas how to make it faster? > > Here is another question: how transaction from secondary storage to primary > works. I see that images on primary are something like delta from original > image (deduplicated?). Which component of Cloudstack is responsible for > transfering data from secondary to primary? > > Thanks in advance for any ideas and info. > > Best regards, > Dmitry > -- Andrija Panić
