Groups.grid.org

Hi,

Some thoughts on EC2 clusters and UC.
Amazon may or may not alter their network configuration to be more cluster friendly out the box. Even if they do it will take a unknown time to happen.

It _appears_ that the Globus Nimbus project's cloudkit offers a way forward.
You could even tag the integrated stack "UniCluster Cumulo" ;)

Specifically, they seem to have an EC2 backend deployed as the "Teraport EC2 Gateway". This seems to have been running EC2 AMIs since late 2007.

So in existing deployments the GLobus-Nimbus front end machine is outside of the EC2.
Nonetheless, it suggests a potential solution if one can get access to the Nimbus EC2 backend code.....?

Once more the two use cases are:
1) Install the Globus-Nimbus components outside of EC2 to create "Your-Own-External EC2 Gateway".
2) Install the Globus-Nimbus components within an EC2 AMI to create "Your-Own-Internal EC2 Gateway" - tautology comes free :)

A reason why 2) might be possible is that the Nimbus pages specifically mention connecting to 'clouds' that require you join their VPN: "Some clouds will only connect your VM to a virtual private network"
See: http://workspace.globus.org/clouds/vpninstructions.html

However, it seems that 1) might even work with a laptop as the gateway machine?
This page (http://workspace.globus.org/clouds/clusters2.html) indicates you can initialize whole clusters, with its own head node, file-node and work nodes. In this context it would seem that the gateway need not play an active part in the cluster!

The benefit of using Nimbus would be tapping into the work done on the one-click cluster start and all the other cloud workspace effort that has been made in the Nimbus project.
See: http://workspace.globus.org/clouds/cloudquickstart.html

Anyway it is early days yet and I won't pretend to fully understand all of Nimbus and the cloudkit.
However, it does seem to me to be the Globus component that is worth working on if the EC2 backend component is released.

Any thoughts on whether this might be the best way to proceed? A sensible way to proceed?

Of course none of this matters until the EC2 backend code is available.... I've dropped the workspace-user mail list a couple of questions. Will update if any info comes to hand.

Regards
Mark