Why I'm reluctant to dive into HPC world, again?

I rarely wrote about my job nowadays. But I think I need a place where I can write about what I feel right now. I can't contain it anymore since I afraid it will burst and affect everyone around me.

Currently I've been tasked to maintain and manage HPC at my workplace. And also to get personnel to utilize it to its full capacity.

Like what I've told most people HPC is totally different from other type of computing. Virtual computing I can still accept. Even cloud computing also I can do. Because both technology only require the dedicated technical team and technology.

For HPC to work there are lots more to consider.

1. The team need to understand what HPC is all about. Also what type HPC that you will built. Is it homogeneous or heterogeneous cluster? Is it built using off-the-shelf commodity hardware or specialized hardware for advanced computation? Is it for life sciences? Weather prediction and forecasting? Wind tunnel simulation? Blueprint creation? Engineering research? Each specialty need a team that know what type of basic software need to be installed in the cluster. Of course the researchers may help but sometimes (sadly to say) their list of software mostly use expensive proprietary software that can't be bought by university's computer center.
2. The software. Most of the software need a different libraries and libraries tools to be installed in the cluster. For life sciences alone you need to install basic libraries for programming languages like C, C++, Java, Perl. For tsunami prediction, you need to install Fortran. Also you need to know which version to install. Some will use Java 1.5. Some will use Java 1.6. All this need to be configured in each and every users inside the HPC. You cannot configure it globally because it will affect the performance and usability of other software. And please bear in mind, the researchers that use the HPC mostly is not IT savvy enough to use all the operating system capabilities.
3. The human factor. The most important thing that usually ignore by everyone in the project. I don't remember which paper state this, but a group of researchers said the most problem encounter by them during HPC projects were human factor. You need to meet with them quite often to get progress from them. That is what I reluctant to do. Actually, I'm not good at communicating with people. And all the person that I need to contact are higher than me in term of rank.

No comments: