Why HPC?

What is HPC?

Are you an academic and a researcher who uses computing to solve problems in science or engineering? Do you ever find yourself running programs on your laptop or desktop that run for hours or even days before they complete and produce results? Do you feel that while you are an expert in your own scientific domain, the use of computers and computing have always been a bit of a struggle for your research workflows? If your answers to the above tend towards "yes" you may have come to the right place to get help with your computational research.

We are in the business of High Performance Computing, or "HPC" for short. HPC is the employment of parallel processing techniques in computer code running on top of specialized hardware in order to solve complex problems "quickly" and on a scale that commodity computing hardware cannot handle. Is it possible that the solution to your computing woes lie in HPC?

Is it for me? How would I know?

Determining whether HPC (or parallel computing) would be useful for your research involves considering several factors:

Nature of Your Research Problem: Assess whether your research involves computationally intensive tasks that can benefit from parallel processing. If your simulations, data analysis, or modeling require significant computational resources and can be broken down into smaller tasks that can run simultaneously, parallel computing might be beneficial.

Size of Data: Consider the volume of data you're working with. If your datasets are large and processing them requires substantial amounts of computational power, storage capacity or throughput, or massive amounts system memory, parallel computing can help distribute the workload across multiple processors or nodes.

Complexity of Algorithms: Evaluate the algorithms you're using in your research. Are they inherently parallelizable? Some algorithms naturally lend themselves to parallel execution, while others may require more sequential processing. For example, compute kernels that rely heavily on matrix multiplications should be suitable for paralellization. Or workflows that rely on Monte Carlo methods also lend themselves easily to paralellization. On the other hand, recursive algorithms that exhibit data dependencies or require global state management may not be inherently parallelizable. For example, algorithms like recursive tree traversal, where each step depends on the results of previous steps, can be challenging to parallelize efficiently without introducing overhead from synchronization or communication between parallel tasks.

Time Constraints: Assess whether time is a critical factor in your research. If you need results quickly and your computations are time-consuming, parallel computing can potentially speed up the process by harnessing multiple processors simultaneously.

Scalability: Determine whether your computational needs are likely to scale in the future. Parallel computing offers scalability by enabling you to add more processing units or nodes as your computational demands increase.

Programming Skills: Assess your programming skills and the feasibility of parallelizing your code. Depending on the programming languages and libraries you're familiar with, parallelizing code may require learning new techniques or using specialized parallel computing frameworks. If required, we can help with that. However, you may also be able to rely on commercial software that simplifies this burden significantly or eliminates it altogether! You might be able to use built-in features in Matlab to implement paralellization for example, or you might (as another example) get Ansys or OpenFOAM to transparently hande the paralellization of your simulation with just a few configuration settings.

By considering these factors, you can make an informed decision about whether high performance computing or parallel computing is suitable for your research needs. Don't be afraid to come talk to us so we can provide insights and guidance tailored to your specific needs. We can figure this out together, and if HPC is for you, we will teach you how to use our system to run your workloads.