[Starcluster] Version of StartCluster with support for Amazon Cluster Compute Images

[Justin Riley]

Hi Fred,

Thanks for this. I've added a link on the homepage and fixed the 
formatting issues for now by surrounding in <code><pre></pre></code> tags.

~Justin

On 09/13/2010 09:50 AM, Fred Rotbart wrote:
> Hi Justin,
>
> I copied the contents to a new wiki page
> http://starcluster.scripts.mit.edu/~starcluster/wiki/index.php?title=StarCluster_Cluster_Compute_AMI_Cookbook_CentOS_5.4
> but the Wiki didn't seem to like the formatting. If it is too messy I
> can send the original as an attachment.
>
> - Fred
>
> On 13/09/10 23:32 , Justin Riley wrote:
>> Fred,
>>
>>> I have built a StarCluster Cluster Compute AMI based on the Rightscale
>>> CentOS hvm image
>>
>> Great work on this. I spent some time previously trying to get Ubuntu migrated to the new HVM image but it turns out that their kernel (2.6.18) will not work with any recent ubuntu distro. I went as far as recreating the HVM's boot and lvm partitions, rsyncing the ubuntu distro, patching the linux-2.6.32 kernel with latest xen pvm-hvm patches, updating the boot/grub/menu.lst for the new kernel but nothing I did seemed to work so I gave up pretty quickly. So in the mean time having a CentOS HVM image is great, thanks for creating and sharing!
>>
>>> On starting up in a StarCluster, the image also exhibits the error
>>> message "mount -t rpc_pipefs sunrpc /var/lib/nfs/rpc_pipefs/" that Dan
>>> Yamins reported in
>>
>> Yea this is the same as my reply to Dan...it's harmless. I'm considering adding "ignore_exit_status=True" to this command since it's executed just to make absolutely sure that the rpc_pipefs is mounted. As you've seen this is not *always* necessary and really should be done by the portmap/NFS startup scripts.
>>
>>> The cookbook is available in
>>> https://docs.google.com/a/dius.com.au/document/pub?id=1ztxOMQZbC5xiCYVRO6IVxqoi81CPdbjxA3FdueyYMsY
>>
>> Also, I can't seem to access your cookbook. Could you please add this to the StarCluster wiki instead? I'd like to have all of the cookbooks centrally available to users. Just copying and pasting it there would be enough, I can handle the formatting issues.
>>
>>> The Cluster Compute cluster seems to work very well without the flaky
>>> mpi behavior that normal Amazon images exhibit.
>>
>> That's excellent, I'm sure that Damian Eads will be excited to hear since he's doing lots of MPI on EC2 last I checked.
>>
>> Thanks again for contributing!
>>
>> ~Justin
>> ________________________________________
>> From: starcluster-bounces at MIT.EDU [starcluster-bounces at MIT.EDU] On Behalf Of Fred Rotbart [rotbart at softbart.com]
>> Sent: Thursday, September 09, 2010 7:04 AM
>> To: starcluster at mit.edu
>> Subject: Re: [Starcluster] Version of StartCluster with support for Amazon Cluster Compute Images
>>
>> I have built a StarCluster Cluster Compute AMI based on the Rightscale
>> CentOS hvm image. It turns out that the Rightscale image did not add
>> anything relevant to StarCluster over the generic Amazon one so I could
>> used that one instead.
>> The image is ami-30b84d59 and has been made public.
>>
>> The image also includes PLAPACK version 3.2 and PETSc 3.1-p4 together
>> with matching mpi4py and petsc4py.
>>
>> On starting up in a StarCluster, the image also exhibits the error
>> message "mount -t rpc_pipefs sunrpc /var/lib/nfs/rpc_pipefs/" that Dan
>> Yamins reported in
>> http://mailman.mit.edu/pipermail/starcluster/2010-August/000337.html.
>> Likewise it does not seem to affect the running of the cluster.
>>
>> The cookbook is available in
>> https://docs.google.com/a/dius.com.au/document/pub?id=1ztxOMQZbC5xiCYVRO6IVxqoi81CPdbjxA3FdueyYMsY
>>
>> Note my cookbook generates full config files rather than the more
>> elegant dynamically editing them.
>>
>> The Cluster Compute cluster seems to work very well without the flaky
>> mpi behavior that normal Amazon images exhibit.
>>
>> - Fred
>>
>> On 31/08/10 17:22 , Fred Rotbart wrote:
>>>      I have cloned the StarCluster git repository to
>>> http://github.com/fcr/StarCluster and added support for the Amazon
>>> Cluster Computer instances.
>>> This has been tested with boto-2.0b2 and with the Amazon generic Cluster
>>> Compute hvm AMI ami-7ea24a17 (which is not, of course, StarCluster ready)
>>>
>>> The "start" command has a new feature that if the -x  option is used, it
>>> will restart a stopped one.
>>> The "stop" command will stop rather than terminate ebs backed instances
>>> (if they are not spot instances).
>>> A new 'terminate" command has been added to actually terminate a running
>>> or stopped cluster.
>>>
>>> Please feel free to try it out and let me know if there are any problems.
>>>
>>> Next challenge is going to be building a StarCluster hvm compatible
>>> image. If anyone has done this, please let know.
>>>
>>> - Fred
>>>
>>
>> _______________________________________________
>> Starcluster mailing list
>> Starcluster at mit.edu
>> http://mailman.mit.edu/mailman/listinfo/starcluster
>>