openstack team mailing list archive

[Yuriy Taraday <yorik.sar@xxxxxxxxx>]

Sorry, my bad.
The more free RAM is better (cost is smaller) thanks to this -1.0. So
the default behavior is spread-first indeed. To make it fill-first,
you can add flag --compute_fill_first_cost_fn_weight=1.0 to scheduler
node.

Kind regards, Yuriy.


On Thu, May 3, 2012 at 8:15 PM, Yuriy Taraday <yorik.sar@xxxxxxxxx> wrote:
> Fill-first cost function returns the amount of free RAM. By default it
> is negated (multiplied by -1.0), so the less free RAM is better.
>
> I think, this is a bit misguiding, but was changed right before Essex
> (see https://bugs.launchpad.net/nova/+bug/965732 ).
>
> Kind regards, Yuriy.
>
>
> On Thu, May 3, 2012 at 8:05 PM, Lorin Hochstein
> <lorin@xxxxxxxxxxxxxxxxxx> wrote:
>> Yuriy:
>>
>>
>> On May 3, 2012, at 4:46 AM, Yuriy Taraday wrote:
>>
>> Just note that since Essex release Nova by default use fill-first cost
>> function, meaning that nodes with less free RAM will be preferred for
>> new instances.
>>
>> Kind regards, Yuriy.
>>
>>
>> I thought the default behavior in essex was spread-first:
>>
>> From:
>>
>> https://github.com/openstack/nova/blob/stable/essex/nova/scheduler/least_cost.py#L41
>>
>>     cfg.FloatOpt('compute_fill_first_cost_fn_weight',
>>              default=-1.0,
>>                help='How much weight to give the fill-first cost function. '
>>                     'A negative value will reverse behavior: '
>>                     'e.g. spread-first'),
>>     ]
>>
>>
>>
>>
>> Take care,
>>
>> Lorin
>> --
>> Lorin Hochstein
>> Lead Architect - Cloud Services
>> Nimbis Services, Inc.
>> www.nimbisservices.com
>>
>>