Re: Parallel filesystems

From: Dustin Machi <no email>
Date: Fri, 1 Jun 2012 13:58:28 -0400

Hey Ron,

Sorry I forgot I was going to post my plugins for this. https://gist.github.com/2853872.

For this plugin, I create the volumes I want to use at amazon and add tags to them. When you tell the gluster cluster to be built, as each node is launched it searches the tags on your AWS volumes to see which it should mount locally. Each volume is given four tags: "brickNode", "type", "Name", and "volume". For a 4 node Gluster cluster, I would create four equally sized volumes. and then assign them like this:

brickNode: 0
type: "glusterbrick"
Name: "myVolume_0"
volume: "myVolume"

For the remaining 3 volumes, they get the same information except the brickNode and Name are incremented.

Assuming these volumes exist already when you launch the data cluster, the plugin will find them, attach them to the instance, format if necessary, and ultimately mount them and create the virtual Gluster volume among all the nodes.

There is probably a lot more detail that i'm leaving out at the moment, but let me know if you have any questions.

Dustin



On May 28, 2012, at 11:58 AM, Rayson Ho wrote:

> The BioTeam also published some tests comparing GlusterFS, local disk,
> & NFS back in 2010:
>
> http://bioteam.net/2010/07/playing-with-nfs-glusterfs-on-amazon-cc1-4xlarge-ec2-instance-types/
>
> Rayson
>
> ================================
> Open Grid Scheduler / Grid Engine
> http://gridscheduler.sourceforge.net/
>
> Scalable Grid Engine Support Program
> http://www.scalablelogic.com/
>
>
>
> On Mon, May 28, 2012 at 10:44 AM, Adam <adamnkraut_at_gmail.com> wrote:
>> Ron,
>>
>> I have a gluster plugin in our repo
>>
>> https://github.com/bioteam/StarClusterPlugins
>>
>> It just creates a distributed volume from the ephemeral disks. It doesn't do
>> anything advanced but it might be enough to get you started.
>>
>> -Adam
>>
>>
>> On Fri, May 25, 2012 at 2:00 PM, Ron Chen <ron_chen_123_at_yahoo.com> wrote:
>>>
>>> That would be great! Can you post them somewhere online?
>>>
>>> -Ron
>>>
>>>
>>>
>>> ----- Original Message -----
>>> From: Dustin Machi <dmachi_at_vbi.vt.edu>
>>> To: Ron Chen <ron_chen_123_at_yahoo.com>
>>> Cc: starcluster <starcluster_at_mit.edu>
>>> Sent: Friday, May 25, 2012 1:54 PM
>>> Subject: Re: [StarCluster] Parallel filesystems
>>>
>>> I'm using glusterfs with my starcluster setup. I have some plugins for it
>>> that I can share with you if you like. They are still works in progress,
>>> but am happy to share them if they are useful.
>>>
>>> Dustin
>>>
>>>
>>>
>>> On May 25, 2012, at 1:25 PM, Ron Chen <ron_chen_123_at_yahoo.com> wrote:
>>>
>>>> Anyone using Lustre, PVFS or GlusterFS on EC2? I'm reading the 'Data
>>>> Sharing Options for Scientific Workflows on Amazon EC2' paper:
>>>>
>>>> http://www.isi.edu/~gideon/publications/JuveG-DataSharing.pdf
>>>>
>>>> Looks like parallel FS like glusterFS can improve cluster performance in
>>>> many cases, especially when there are more than 8 nodes.
>>>>
>>>> -Ron
>>>>
>>>> _______________________________________________
>>>> 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
>>
>>
>>
>> _______________________________________________
>> StarCluster mailing list
>> StarCluster_at_mit.edu
>> http://mailman.mit.edu/mailman/listinfo/starcluster
>>
>
>
>
> --
> ==================================================
> Open Grid Scheduler - The Official Open Source Grid Engine
> http://gridscheduler.sourceforge.net/
>
> _______________________________________________
> StarCluster mailing list
> StarCluster_at_mit.edu
> http://mailman.mit.edu/mailman/listinfo/starcluster
Received on Fri Jun 01 2012 - 13:58:34 EDT