Open Stack

On Sep 1, 2016 4:53 PM, "Leon Liu" <leon.liu at viscore.com> wrote:
>
> Hi All,
>
>
>
> We have read from others that the dynamic and large number VM
provisioning might be a pain, e.g.
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6719385 The scenario is
that the number of instances for any application is continuously changing,
and the number is as large as hundreds of instances in hundreds of nodes.
>
>
>
> However, with our limited investigation, we still see that it takes 200
seconds to 700 seconds for OpenStack to launch 100 instances. So, we wonder
if large scale provisioning and the time involved could translate into a
true pain for Datacenter operators? If it is, perhaps OpenStack already
addresses this but we are just not aware.
>
>
I grant you I have never tried this many at once, but I can't imagine a
properly sized Openstack cluster taking anywhere near that long to
provision 100 instances assuming that:

1) You have an appropriate number of Nova Scheduler and Nova API workers

2) A sufficient number of compute nodes

3) Enabled base instance caching on the hypervisor or ephemeral storage in
a shared backend. We already have the capability to store glance images on
the hypervisor or to use Ceph copy-on-write with RBD ephemeral storage.

>
> Would a more effective way of VM provisioning, if integrated into
OpenStack, be of any value?
>
>
Perhaps for non-kvm or xen shops. You would need to benchmark against the
above available options to know for sure.
>
> Best Regards
>
> Leon Liu
>
> Viscore Technologies Inc.
>
> http://www.viscore.com
>
>
> _______________________________________________
> Mailing list:
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>

Cheers,
Erik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20160901/34758299/attachment.html>