[Openstack] Which nova scheduler for different hardware sizes?

Everett Toews everett.toews at cybera.ca
Tue Dec 13 23:13:02 UTC 2011


I know I'm reviving an older thread here but we implemented our own simple
"hardware aware" scheduler for Cactus. Maybe it will help you?

https://github.com/canarie/dair/blob/master/OpenStack/misc/hardware.py

Everett

On Thu, Nov 10, 2011 at 11:35 AM, Christian Wittwer <wittwerch at gmail.com>wrote:

> Thanks a lot for your answers. Unfortunately I can't follow the trunk
> and I have to use the Diablo release. Is it possible to backport that
> new scheduler to Diablo?
> Anyway I gave the least cost scheduler a try, it loads but never
> schedules a vm correctly.
>
> --compute_scheduler_driver=nova.scheduler.least_cost.LeastCostScheduler
>
> But then log fires that message when I try to launch an instance
>
> 2011-11-10 19:31:15,588 DEBUG nova.scheduler.least_cost [-] Weighted
> Costs => [] from (pid=1697) weigh_hosts
> /usr/lib/pymodules/python2.7/nova/scheduler/least_cost.py:170
> 2011-11-10 19:31:15,591 ERROR nova.rpc [-] Exception during message
> handling
> (nova.rpc): TRACE: Traceback (most recent call last):
> (nova.rpc): TRACE:   File
> "/usr/lib/pymodules/python2.7/nova/rpc/impl_kombu.py", line 620, in
> _process_data
> (nova.rpc): TRACE:     rval = node_func(context=ctxt, **node_args)
> (nova.rpc): TRACE:   File
> "/usr/lib/pymodules/python2.7/nova/scheduler/manager.py", line 103, in
> _schedule
> (nova.rpc): TRACE:     host = real_meth(*args, **kwargs)
> (nova.rpc): TRACE:   File
> "/usr/lib/pymodules/python2.7/nova/scheduler/abstract_scheduler.py",
> line 231, in schedule_run_instance
> (nova.rpc): TRACE:     raise driver.NoValidHost(_('No hosts were
> available'))
> (nova.rpc): TRACE: NoValidHost: No hosts were available
> (nova.rpc): TRACE:
>
> My compute hosts are up and running, and there are no other instances
> running.
>
> root at unic-dev-os-controller:~# nova-manage service list
> Binary           Host                                 Zone
> Status     State Updated_At
> nova-compute     unic-dev-os-controller               nova
> enabled    :-)   2011-11-10 18:35:06
> nova-scheduler   unic-dev-os-controller               nova
> enabled    :-)   2011-11-10 18:35:06
> nova-network     unic-dev-os-controller               nova
> enabled    :-)   2011-11-10 18:35:06
> nova-compute     unic-dev-os-compute1                 nova
> enabled    :-)   2011-11-10 18:35:06
>
> Any ideas why the scheduler does not find a valid host?
>
> Christian
>
> 2011/11/1 Lorin Hochstein <lorin at isi.edu>:
> > Christian:
> >
> > Sandy's branch just landed in the repository. You should be able to use
> the distributed scheduler with the least cost functionality by specifying
> the following flag in nova.conf for the nova-scheduler service:
> >
> >
> --compute_scheduler_driver=nova.scheduler.distributed_scheduler.DistributedScheduler
> >
> > By default, this uses the
> nova.scheduler.least_cost.compute_fill_first_cost_fn weighting function.
> >
> > Note, however, that this function will favor scheduling instances to
> nodes that have the smallest amount of RAM available that can still fit the
> instance. If you're looking for the opposite effect (deploy to the node
> that has the most amount of RAM free), then you'll have to write your own
> cost function.  One way would be to add the following method to
> least_cost.py:
> >
> >
> > def compute_least_loaded_cost_fn(host_info):
> >    return -compute_fill_first_cost_fn(host_info)
> >
> >
> > Then add the following flag to your nova.conf
> >
> >
> --least_cost_functions=nova.scheduler.least_cost.compute_least_loaded_cost_fn
> >
> >
> > Lorin
> > --
> > Lorin Hochstein, Computer Scientist
> > USC Information Sciences Institute
> > 703.812.3710
> > http://www.east.isi.edu/~lorin
> >
> >
> >
> >
> > On Nov 1, 2011, at 11:37 AM, Sandy Walsh wrote:
> >
> >> I'm hoping to land this branch asap.
> >> https://review.openstack.org/#change,1192
> >>
> >> It replaces all the "kind of alike" schedulers with a single
> DistributedScheduler.
> >>
> >> -S
> >>
> >> ________________________________________
> >> From: openstack-bounces+sandy.walsh=rackspace.com at lists.launchpad.net[openstack-bounces+sandy.walsh=
> rackspace.com at lists.launchpad.net] on behalf of Christian Wittwer [
> wittwerch at gmail.com]
> >> Sent: Tuesday, November 01, 2011 5:38 AM
> >> To: Lorin Hochstein
> >> Cc: openstack at lists.launchpad.net
> >> Subject: Re: [Openstack] Which nova scheduler for different hardware
> sizes?
> >>
> >> Lorin,
> >> Thanks for your reply. Well the least cost scheduler with these cost
> >> functions looks interesting.
> >> Unfortunately there is not much documenation about it. Can somebody
> >> give me an example how to switch to that scheduler using the memory
> >> cost function which already exist?
> >>
> >> Cheers,
> >> Christian
> >>
> >> 2011/10/24 Lorin Hochstein <lorin at isi.edu>:
> >>> Christian:
> >>> You could use the least cost scheduler, but I think you'd have to
> write your
> >>> own cost function to take into account the different number of cores.
> >>> Looking at the source, the only cost function it comes with only takes
> into
> >>> account the amount of memory that's free, not loading in terms of total
> >>> physical cores and allocated virtual cores. (We use a custom scheduler
> at
> >>> our site, so I don't have any firsthand experience with the least-cost
> >>> scheduler).
> >>> Lorin
> >>> --
> >>> Lorin Hochstein, Computer Scientist
> >>> USC Information Sciences Institute
> >>> 703.812.3710
> >>> http://www.east.isi.edu/~lorin
> >>>
> >>>
> >>>
> >>> On Oct 22, 2011, at 3:17 AM, Christian Wittwer wrote:
> >>>
> >>> I'm planning to build a openstack nova installation with older
> >>> hardware. These servers obviously doesn't have the same hardware
> >>> configuration like memory and cores.
> >>> It ranges from 2 core and 4GB memory to 16 core and 64GB memory. I
> >>> know that there are different scheduler, but I'm not sure which one to
> >>> choose.
> >>> The simple scheduler tries to find the least used host, but the amount
> >>> of used cores per host (max_cores) is a constant, which doesn't work
> >>> for me.
> >>> Maybe the least cost scheduler would be the right one? But I'm not
> >>> sure, because I did not find any documenation about how to use it.
> >>>
> >>> Cheers,
> >>> Christian
> >>>
> >>> _______________________________________________
> >>> Mailing list: https://launchpad.net/~openstack
> >>> Post to     : openstack at lists.launchpad.net
> >>> Unsubscribe : https://launchpad.net/~openstack
> >>> More help   : https://help.launchpad.net/ListHelp
> >>>
> >>>
> >>
> >> _______________________________________________
> >> Mailing list: https://launchpad.net/~openstack
> >> Post to     : openstack at lists.launchpad.net
> >> Unsubscribe : https://launchpad.net/~openstack
> >> More help   : https://help.launchpad.net/ListHelp
> >
> >
>
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack at lists.launchpad.net
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20111213/9ac743f2/attachment.html>


More information about the Openstack mailing list