Choose timezone
Your profile timezone:
Powered by Indico
v3.3.2
Over the last few years, HPC centers have observed an increasing demand and usage of data-intensive projects. These workloads have a measurable impact on the provided storage systems which has led to problems in the past due to a "noisy-neighbour" effect. This means, that for example one or just a few users doing small file I/O, e.g., by running a machine learning task can impact the performance of all other users' experience. Similarly, the actual storage performance becomes a key factor in the overall performance of these users.
Other related challenges with regard to data-intensive projects are efficient data transfers since these methods usually require large input data sets. Similarly, data sharing becomes an important aspect of fostering collaborations. Lastly, also novel interaction paradigms become increasingly important in attracting a new user group to HPC systems. These users do not want to work with an SSH shell but instead, look for more user-friendly and accessible ways to interact with an HPC system and manage their jobs and data.
This workshop, which is held within the context of the "Large Scale Data Management" future project ("Zukunftsprojekt") of the NHR, explores all of these aspects through a number of contributed talks. First, the four involved NHR centers will give an update on their storage strategies. Within a second block of contributed talks, a wide range of topics related to the support of data-intensive projects on HPC is provided to give an overarching overview which will serve as a basis for a concluding discussion and further collaborations.
This workshop will provide interesting insides for admins of HPC systems, and users alike and everyone is welcome to join. It will be held online and also a selected, partial participation is possible.
We gratefully acknowledge funding received by the Nationales Hochleistungsrechnen by the strategic project “Large Scale Data Management”