Do you have sensitive data?
By sensitive data we mean data that is either red or black according to the data classification guide. If you have black data that you want to use in your simulations, you have to use Services for sensitive data (TSD). If you have red data, you can use either TSD or Educloud Research.
Have you checked with local IT in your own unit if you have any existing special IT solutions?
Several local units have special IT solutions tailored to serve the needs of your faculty, department or center. Have you checked with local IT in your unit if there are any solutions that might cover your needs?
Are you going to collaborate with external researchers?
If you are going to collaborate with researchers who do not work at UiO, the easiest option is to use Educloud Research. Through Educloud Research you can also get access to the supercomputer Fox. Educloud also provides server space for storing data, and you can administer your project and access levels for collaborators yourself. Educloud research is approved for green, yellow and red data.
How comfortable are you with working in a console?
Many of the compute resources available through UiO require some familiarity with IT, especially to get started. If you need a more user friendly service, you should specify this when you search for a solution. However, the more advanced your computational needs are, the more likely it is that the solution you need requires some advanced knowledge of IT. In such cases we recommend that you contact your local IT or an IT technician affiliated with your unit (research group, department, faculty) for help getting started.
Do you primarily need access to more software than you have installed on your own computer?
If you primarily need access to more software to run analyses, we recommend using the Statistics servers, available through UiO Programkiosk. Log in and choose "Statistikk Fullskjerm".
Do you need much compute power?
We use four different categories of computational power needs: light, medium, heavy and very heavy computations. In order to run your analysis on a High Performance Computation (HPC) cluster does not mean that you get access to a computer that is faster than the one you have yourself, but that you get access to more cores at the same time. This is a benefit if you can run different parts of the analysis on different cores in parallel at the same time, which means that the complete analysis will be faster. If you can not parallelise your code, running it on a HPC cluster will not be any faster than running it on your own computer. However, you do get the benefit that the analysis is not taking up resources on your computer, allowing you to use your own machine for other processes while the analysis is running.
Light weight
If you need to run analyses or simulations that only take from a few minutes to a few hours, and you wish to be able to also use your own computer for other things at the same time, we recommend you use the statistics servers. You can read more about statistics servers here.
Medium
This category is for analyses or simulations that would take a few days on your own machine. It this level you might want to consider if your code can be parallelised, so that you can run it on multiple cores at the same time. For this level there are multiple resources that might be appropriate.
Heavy
In this category your computations can take several weeks to run on your own computer. If at all possible you should parallelise your code, and you will need access to several cores at the same time. For this level there are multiple resources that might be appropriate.
Very heavy
This category covers needs where even though your code is parallelised and run on multiple cores, you still need days or weeks to run.
If you are in doubt whether your computation needs fall within the categories heavy or very heavy, we strongly recommend that you try heavy first.
AI/ML
If you are going to specifically work with machine learning and training models, we also have dedicated resources tailored for this. These servers have more GPUs and software specifically for ML. (Light ML: the ML-nodes, medium ML: Fox).
How much data do you have?
The different compute resources have access to different options for storing data, with different disc sizes. Because of this it is useful to know roughly how much data you have or will have over time. Be aware that working on large numbers of individual files also requires significant memory from the resource that you will be using.
We use three different categories of data quantity: 0-100 GB, 100 GB-10 TB, or more than 10 TB.
0-100 GB
Typically this will be most datasets consisting of primarily text, numbers, spreadsheets, survey responses (e.g. Nettskjema data), etc. 1 GB of pure text corresponds to about 678000 pages, or a Word document of about 65000 pages.
If you have a very large number (millions) of small files that in total add up to less than 100 GB, you might consider choosing a compute resource that is adapted for the next category (100 GB-10 TB).
100 GB-10 TB
If you work a lot with sound or video you may easily go over the 100 GB limit. The same goes for image data and 3D scans. 100 GB corresponds to about 200 hours of standard quality video, or 50 hours of high quality video.
>10 TB
Very few fall within this category, but some research projects definitely do (you know who you are).
Where is your data stored now, and can they be moved?
Several of our compute resources require that your data is stored in a specific location in order to be accessible. Because of this it is important that you figure out where your data is or will be stored, and if it will be possible to move them. Especially with large data it can be challenging to move them to a new location. In some cases we will be able to help you with this (contact us: itf-ft@usit.uio.no).
How sure are you that your current needs will not change dramatically over time?
Changing from one compute resource to another can be quite challenging. Therefore you should not just consider your needs right now, but what your needs might be in the near future. If you know that your needs will change significantly later on, and that you will either have much more data or will need to run significantly more heavy computations, you should try and find a solution that also serves your future needs.
Special solutions
Do you need a solution tailored to your needs specifically? We can help set up special servers for most types of projects. Contact us to figure out a solution: itf-ft@usit.uio.no