Re: Terminating a cluster does not remove its security group (UNCLASSIFIED)

From: Oppe, Thomas C ERDC-RDE-ITL-MS Contractor <no email>
Date: Tue, 18 Dec 2012 18:24:33 +0000

Classification: UNCLASSIFIED
Caveats: FOUO

Dustin,

Thank you for the idea of using the local (ephemeral) disk associated with a particular instance, say the "master" node. In our case, we are benchmarking HYCOM, an ocean-modeling code. It is possible to generate two kinds of executables: a serial I/O (sio) executable in which all I/O is routed through MPI rank 0 (i.e., master) and a parallel I/O (pio) executable that uses MPI-2 I/O calls. Hopefully the master instance has a 200GB local disk because that is how much I/O is generated during a run (I think it is actually 150 GB or so). I will see if there is any kind of instance in StarCluster that has this size of local disk.

Thank you for the information.

Tom Oppe
 

-----Original Message-----
From: Dustin Machi [mailto:dmachi_at_vbi.vt.edu]
Sent: Tuesday, December 18, 2012 12:04 PM
To: Oppe, Thomas C ERDC-RDE-ITL-MS Contractor
Cc: starcluster_at_mit.edu
Subject: Re: [StarCluster] Terminating a cluster does not remove its security group (UNCLASSIFIED)


> On another issue, do you have a pointer to information about how to
> create and use striped disks within StarCluster? We have an
> application that is I/O-intensive and the I/O portion seems to be
> slowing down the execution time considerably.

The I/O is being done over NFS (over the local network) by default. I don't think there is anything built in that will let you configure a striped setup out of the box. It would likely require that you write a plugin to do that job. Your plugin would need to create the volumes, mount them locally, and then configure and mount the logical (striped) volume on top of that and format it with XFS. This XFS volume is then exported to the other nodes by NFS.

All that said, I'm not sure NFS will cut it for you with a lot of I/O.
I would suggest that you use the local TEMP disk (ephemeral disks) on your compute jobs and then copy the results over when a particular work set is complete. This allows your app to use local/non-shared disk for the intensive stuff and then just share back afterwards.

Dustin

Classification: UNCLASSIFIED
Caveats: FOUO
Received on Tue Dec 18 2012 - 13:24:37 EST