Re: Smart Cluster

From: Ryan Golhar <no email>
Date: Tue, 29 Jan 2013 11:07:04 -0500

I propose this question/answer be added added to a FAQ.


On Tue, Jan 29, 2013 at 9:31 AM, MacMullan, Hugh
<hughmac_at_wharton.upenn.edu>wrote:

> Hi Sergio:
>
> a -- yes!
>
> MASTER_INSTANCE_TYPE = c1.medium
> MASTER_IMAGE_ID = ami-XXXXXX
> NODE_IMAGE_ID = ami-XXXXXX
> NODE_INSTANCE_TYPE = cr1.8xlarge
>
> b -- in config do:
>
> In the [cluster NAME] section:
> DISABLE_QUEUE=True
> PLUGINS = sge
>
> Then define your sge plugin with:
>
> [plugin sge]
> setup_class = starcluster.plugins.sge.SGEPlugin
> master_is_exec_host = False
>
> You can also force the master to be a spot instance by using 'starcluster
> start --force-spot-master', which can save serious cash (90%).
>
> -Hugh
>
>
>
> On Jan 29, 2013, at 9:02 AM, "Sergio Mafra" <sergiohmafra_at_gmail.com>
> wrote:
>
> Hi fellows,
>
> If we are supposed to have the instances running for a long time,
> LoadBlancer can help us a lot, turning on and off when is needed.
> I wonder if we can make this more smarter and cheaper, launching the
> master node as a m1.small instance, taking care only for communication and
> queue managemente, leaving the heavier job for the compute nodes, that can
> take advantage of cluster type instance and placement groups.
> So my question is:
>
> a) Can we have a cluster with differente instance types - Master and
> Compute nodes?
> b) How to exclude Master node from being a computacional node in the job
> queue?
>
>
> All the best,
>
> Sérgio
>
> _______________________________________________
> StarCluster mailing list
> StarCluster_at_mit.edu
> http://mailman.mit.edu/mailman/listinfo/starcluster
>
>
> _______________________________________________
> StarCluster mailing list
> StarCluster_at_mit.edu
> http://mailman.mit.edu/mailman/listinfo/starcluster
>
>
Received on Tue Jan 29 2013 - 11:07:06 EST