"Beyond FLOPS: Achieving High Productivity on a Linux Cluster System"

 

As Linux clusters have matured, the criteria measuring their success has evolved.  Rather than measuring a system’s effectiveness with a Linpack benchmark or placement on the Top 500 list, systems are now measured by the productivity they deliver to the user community they serve.  Once the machine is installed the work is not complete.  Several areas must be addressed on a continual basis to increase the use-ability and utilization of a Linux cluster.

 

This presentation will center on the steps taken by an actual customer to achieve increased productivity on their Linux cluster system.  Both successes and failures associated with each of the following topics will be addressed:

 

-          System architecture and optimization

-          Supporting software and tools

-          Hardware and software system maintenance

-          Porting/optimizing end-user applications

-          Education of system administrators and end-users

 

Since Linux clusters are a conglomeration of many third party components, vendor involvement with both customers and third party providers is absolutely essential to deploying and running production systems effectively.   Key aspects of vendor involvement will be addressed as to how they relate to a customer achieving the highest productivity on their system.