Possible Vagrant configuration change

[Mark Chu-Carroll <mc...@apache.org>]

Folks:

I've been working on getting together an and-to-end test for open-source
Aurora. In order to get it working with a realistic scenario including
multiple task instances, I've modified the vagrant configuration to launch
to mesos slaves, each of which believes that it has 4cpus to allocate. I'd
like to make this two-slave setup the default vagrant configuration.

On my work laptop (Macbook Pro, 4 cores, 8GB memory), it's
indistinguishable from running with just one slave - there's no visible
impact on the performance of other tasks, and running little benchmarks on
my laptop show no statistically meaningful variation between the 1 and 2
slave configs when no tasks are running on the slaves.  I suspect that with
less memory, there would be more performance impact, but I can't measure
that.

Does anyone object to this change?  (If so, I'll put this as a new
Vagrantfile in the directory containing the end-to-end tests. But I'd
really prefer to have a single canonical vagrant test/demonstration
configuration.)

Thanks,

        -Mark

Re: Possible Vagrant configuration change

[Chris Lambert <cl...@twitter.com>]

Seems more realistic.  +1

Is there any documentation that might depend on this?



On Sat, Feb 1, 2014 at 6:06 PM, Mark Chu-Carroll <mc...@apache.org>wrote:

> Folks:
>
> I've been working on getting together an and-to-end test for open-source
> Aurora. In order to get it working with a realistic scenario including
> multiple task instances, I've modified the vagrant configuration to launch
> to mesos slaves, each of which believes that it has 4cpus to allocate. I'd
> like to make this two-slave setup the default vagrant configuration.
>
> On my work laptop (Macbook Pro, 4 cores, 8GB memory), it's
> indistinguishable from running with just one slave - there's no visible
> impact on the performance of other tasks, and running little benchmarks on
> my laptop show no statistically meaningful variation between the 1 and 2
> slave configs when no tasks are running on the slaves.  I suspect that with
> less memory, there would be more performance impact, but I can't measure
> that.
>
> Does anyone object to this change?  (If so, I'll put this as a new
> Vagrantfile in the directory containing the end-to-end tests. But I'd
> really prefer to have a single canonical vagrant test/demonstration
> configuration.)
>
> Thanks,
>
>         -Mark
>

Re: Possible Vagrant configuration change

[Mark Chu-Carroll <mc...@apache.org>]

Sorry - the review is in https://reviews.apache.org/r/17457/.

   -mark


On Mon, Feb 3, 2014 at 2:23 PM, Bill Farner <wf...@apache.org> wrote:

> Can you link to the review?  I can pull down the patch and try it out.
>
> -=Bill
>
>
> On Sat, Feb 1, 2014 at 6:06 PM, Mark Chu-Carroll <mchucarroll@apache.org
> >wrote:
>
> > Folks:
> >
> > I've been working on getting together an and-to-end test for open-source
> > Aurora. In order to get it working with a realistic scenario including
> > multiple task instances, I've modified the vagrant configuration to
> launch
> > to mesos slaves, each of which believes that it has 4cpus to allocate.
> I'd
> > like to make this two-slave setup the default vagrant configuration.
> >
> > On my work laptop (Macbook Pro, 4 cores, 8GB memory), it's
> > indistinguishable from running with just one slave - there's no visible
> > impact on the performance of other tasks, and running little benchmarks
> on
> > my laptop show no statistically meaningful variation between the 1 and 2
> > slave configs when no tasks are running on the slaves.  I suspect that
> with
> > less memory, there would be more performance impact, but I can't measure
> > that.
> >
> > Does anyone object to this change?  (If so, I'll put this as a new
> > Vagrantfile in the directory containing the end-to-end tests. But I'd
> > really prefer to have a single canonical vagrant test/demonstration
> > configuration.)
> >
> > Thanks,
> >
> >         -Mark
> >
>

Re: Possible Vagrant configuration change

[Bill Farner <wf...@apache.org>]

Can you link to the review?  I can pull down the patch and try it out.

-=Bill


On Sat, Feb 1, 2014 at 6:06 PM, Mark Chu-Carroll <mc...@apache.org>wrote:

> Folks:
>
> I've been working on getting together an and-to-end test for open-source
> Aurora. In order to get it working with a realistic scenario including
> multiple task instances, I've modified the vagrant configuration to launch
> to mesos slaves, each of which believes that it has 4cpus to allocate. I'd
> like to make this two-slave setup the default vagrant configuration.
>
> On my work laptop (Macbook Pro, 4 cores, 8GB memory), it's
> indistinguishable from running with just one slave - there's no visible
> impact on the performance of other tasks, and running little benchmarks on
> my laptop show no statistically meaningful variation between the 1 and 2
> slave configs when no tasks are running on the slaves.  I suspect that with
> less memory, there would be more performance impact, but I can't measure
> that.
>
> Does anyone object to this change?  (If so, I'll put this as a new
> Vagrantfile in the directory containing the end-to-end tests. But I'd
> really prefer to have a single canonical vagrant test/demonstration
> configuration.)
>
> Thanks,
>
>         -Mark
>