I have just compiled a short guide on how to build the exodusII library.
The exodusII library is an open CAE filetype that has the ability to read and write “parallel” files. In terms of HPC computing, this means that files can be written out in a much more efficient manner. I am keeping the actual process secret for now, but the time taken between solver starting to write results and postprocesser finishing reading results looks like it will be reduced by about 100 times!
This isn’t really a “tutorial”, more of a “how to” guide, but it should provide assistance to anyone trying to get exodusII compiled.
The old site hosted internally at bc247.eng.monash.edu.au has been migrated across to a wordpress.com site. This means that this site is once again accessible to the outside world, and hopefully the information contained will be useful to all. It also means that the bc247 head node no longer needs to run apache, so one of the hundreds of tasks running on it can be switched off.
So apparently you SHOULD NOT COMPILE Kernels with the Ring Buffer Stress Tester (Kernel Hacking > Tracers). This causes the executables rb_producer and rb_consumer to be created and run all the time. These are testing functions that use 100% of available CPU. Without these running, there should be a marked speedup in performance.
The kernel is still version 3.7.4, but rebuilt to be more efficient.
Ansys 14.5 is now on Beowulf. The entire suite is installed including ANSYS Mechanical (APDL), Fluent, CFX, ICEMCFD and E-mag Products
After some investigation work, it was found that the script to reboot the cluster nodes did not have enough time delay between each node reboot to stop the network from over-congesting. As a result, was not able to respond to all DHCP requests from the nodes in a timely manner, and not all nodes were rebooting back into the cluster. The node reboot script has therefore been updated with an increased delay between each node rebooting to ensure that the network does not become over-congested during node rebooting, so that all DHCP requests are answered reliably by the head node.
We are now likable using Facebook, Google+, Twitter and LinkedIn. Just click on one of the buttons to show your support.
There was also a button for Xing, but I have no idea what Xing is, so I got rid of it. If you are a Xing user, please comment so I can reactivate the button, and then you can Xing Beowulf
A new page has been added to the site to explain the complicated node boot process (link). The page is defiantly not yet complete, only the first few stages have been fully explained. It is hoped that this page may serve as a reference to other cluster builders.
By offering all this information for free, I am hoping to inspire a new generation of clusterers. People of the world, put down you iPad, iPhone, iMac, iPod and learn to do something useful with a computer. Stop wasting your time making photo montages, and go out a build a cluster! Only once you can solve complicated engineering and scientific problems, can you really experience the true exhilaration of life! </inspirational>