Partner Newsletter Q2 2010 - Intel® Cluster Ready Articles 4


Engineering and Manufacturing Use Cases for Intel® Cluster Checker 1.5's Auto Configuration Feature

By Christopher Heller

Hello, Intel Cluster Ready partners! This is the first of the technical articles you will come to count on from the Intel Cluster Ready quarterly newsletter. We want these articles to aid your Engineering and Manufacturing efforts for your Intel Cluster Ready implementations. This first article provides practical tips on how to use the auto configuration feature present in Intel Cluster Checker 1.5.

Start by downloading Intel Cluster Checker 1.5, the Intel Cluster 3.0 runtimes (recommended), and the licenses for both products. I personally recommend creating a new XML file with the release of Intel Cluster Checker 1.5; there are a lot of new features like global configurations to take advantage of.

Next, make sure to source Intel Cluster Checker into your environment. You can do this manually via a command such as source /etc/intel/clck/1.5/ or via a user's environent (bashrc, for example) or globally via /etc/bashrc (for example).

Now, let's use one of the example XML configuration files present in /opt/intel/clck/1.5/examples. I suggest using the example-auto.xml file to start. Make sure you rename the file to something you can remember and will want to keep as a reference for your future Cluster Checker runs and designs. If the node list is not specified in a default location such as /etc/intel/clck/nodelist, you'll need to either create the nodelist in a default location, or change the <nodefile> flag to point to where your nodelist is located. My suggestion is to place/create your Cluster Checker nodefiles in default locations. If you're not sure where the node list file is located, run the tool with the --auto flag; if there is no nodefile in that location, Cluster Checker will tell you.

Now for the fun part. Making sure you're running as root, to ensure the correct values for all applicable tests are set into the xml file, run something like this:

'$cluster-check ./Westmere-auto.xml --auto' (example XML file)

Intel Cluster Checker should now run through quite a few tests. Runtimes vary from 2-15 minutes for a four-node cluster. If your cluster takes longer, that is not an instant indication that something is wrong. Be patient and go back later to check the results. If the code is running on a truly homogeneous cluster, errors will most likely be minimal or nonexistent. Autoconfiguration does some grouping of nodes, but is not comprehensive, so on a heterogeneous cluster, you'll need to make some modifications to the nodefile and XML configuration file to let Intel Cluster Checker know how to deal with the different groups of nodes your cluster may have. For more information about setting up node groups, see the Groups section in the Mass-Producing Your Certified Cluster Solutions document.

Once the run has completed, the modified XML from the run will be present in the XML file used for the test. You'll also notice that Cluster Checker has found the runtimes and placed them into <global_configuration>.

Now, look at the modified XML file and make sure the performance numbers look like they're in the ball park. For Engineering, setting performance guidelines is very important for implementation in Manufacturing. Each company needs to decide if the threshold set from auto configuration is acceptable (internally) for performance expectations for its own product(s). If the value is acceptable for mass production in Manufacturing, then on every cluster, you can use the --auto or --auto-configuration feature to set single-node performance values, saving valuable time and effort compared to calculating these thresholds by hand. The resulting XML configuration file can be used in --deployment final checks before the cluster goes out the door. In addition, your end-users can use the file with Intel Cluster Checker to check the cluster or as a support aid for your company to your customer.

You can also read the Intel Cluster Checker User's Guide and man page to learn about several options to break down the features of Intel Cluster Checker 1.5's auto configuration feature.

This article has described the most common usage scenario where all features will be used. I hope you will enjoy this new feature and provide us with feedback on your experiences for future improvements.

Christopher Heller is an HPC customer support and development engineer within the Intel Cluster Ready program. He's based in Champaign, IL and has worked at Intel for 5 years.

We want to hear your feedback. Leave us a comment or question below about the article(s), and/or please suggest topics you would like to read about in future newsletters.

Before posting a comment, please click here to login. If you are a new user, click here to register.


For more complete information about compiler optimizations, see our Optimization Notice.