April 15th, 2009 11:54 am
Posted by Brock Taylor
Tags: blog, HPC, ICR, Intel Cluster Ready, OFED, one-to-many, SMB
Intel® Cluster Ready (ICR) solutions aren't meant to be unique works of art painstakingly hand-crafted for a single deployment. Instead, ICR clusters are designed to allow a mix-and-match of solutions and applications. We usually refer to this as the "one-to-many" aspect that is a basic principle of the program. This is my attempt to explain what that means.
There are really two "one-to-many" viewpoints in ICR, but I’m going to focus on the benefit of a single ICR solution that is capable of executing any of the registered applications out of the box. By defining a minimal software interface, ICR enables a solutions vendor to design and certify a cluster without a prior knowledge of the specific applications that will run on the solution. Separately, HPC software vendors register and demonstrate that their applications execute on top of the ICR defined interface. That means the applications now map to the solution interface with the underlying mechanisms of the cluster being largely transparent. As long as cluster provides the ICR interface, any registered application is capable of running. One solution, many applications.
So what does this really mean? For the newbie cluster administrator or the would-be cluster user, it means the barrier or knowledge ramp has come way down. ICR enables using clusters without having to learn steps like installing OFED from source or building a custom compute node image. This can be especially valuable to small or medium-sized companies looking to clusters for competitive advantage but find the expertise ramp too steep. You don't want the top scientist at your company spending time learning how to build clusters; you want that person spending time utilizing the cluster to get better, quicker results. ICR and the "one-to-many" approach enables moving to clusters without the expensive learning curve.
For the experienced cluster administrator or IT manager, ICR's "one-to-many" approach has additional benefits. The solution comes ready to run registered applications, so there's a lot less work required to deploy and to start using the resource. Registered applications get installed, and users start executing. It's a faster ROI on the new infrastructure. Adding more registered applications after the solution is deployed is also less work because the solution doesn't require rebuilding - the required software interface is already there and verified. It adds up to less time spent launching and maintaining new infrastructure and more focus on getting the right amount of resources to meets the needs of the users.
If there's a simple meaning for "one-to-many," though, I'd say it means confidence in the cluster. The solutions vendors are able to create processes that produce the same cluster every time which builds continuity and consistency in the solution, and the application vendors are able to rely on an interface that is verified to be present. A cluster solution can easily be purchased and deployed for a range of applications without having to understand the relation to the system libraries, required utilities, or middleware components. The cluster becomes more like an appliance used to tackle complex problems, and that makes life easier and simpler whether you are a cluster newbie or a guru.
If you want to see this principle in action, we teamed with ICR partners last year to showcase "one-to-many" solutions. Here's one account of those proof-point events: Accelerating HPC Productivity with Intel® Cluster Ready
JOIN THE CONVERSATION
You must be a Registered Member in order to comment on Cluster Connection posts.
Members enjoy the ability to take an active role in the conversations that are shaping the HPC community. Members can participate in forum discussions and post comments to a wide range of HPC-related topics. Share your challenges, insights and ideas right now.
Login Register Now