Computational facilities @ MPA
The MPA has always placed considerable emphasis on computational astrophysics and has therefore ensured access to forefront computing facilities both in-house and externally.
The current in-house system is based on a large number of central LINUX-based compute-, file-, and network-servers. It is operated by MPA's own IT-group. Users have free access to all workstations and are connected via Linux desktop-PCs or via their laptops over VPN. Data is stored on central fileservers with a capacity of several hundred Terabyte, and distributed with the filesystems AFS and GPFS. All critical data is backed-up daily.
In addition to the desktop systems, which amount to more than 150 fully equipped workplaces, users have access to central number crunchers. The compute cluster currently offers about 3000 cores. The total on-line disk capacity at MPA is at the Petabyte level; individual users control disk space ranging from a mere GB to dozens of TB, according to scientific need. Energy consumption and cooling have become a crucial aspect of IT-installations. At MPA, we are concentrating on low power-consumption hardware and efficient, environmental-friendly cooling.
For larger computing tasks, MPA scientists use the facilities of the central computing centre of the Max-Planck Society (known as MPCDF), which is located a few hundred metres from the MPA, as well as the nearby Leibniz Computer Centre of the state of Bavaria (the LRZ) and other German supercomputer centres (e.g. in Stuttgart and Jülich). Facilities at the MPCDF include massive parallel supercomputers, a large number of mid-range compute-clusters, of which one is owned and exclusively used by MPA together with a mass storage system with a capacity in the multi-PB range. This is the Freya cluster.
MPA-cluster Freya @ MPCDF
~8000 cores, 60 high-performance GPUs, 43TB core memory, 4.5PB storage
Freya is used for analysis, small to medium simulations, and for developing parallel codes. In addition, MPA operates a core node of the Virgo (the “Virgo supercomputer consortium”) data center at the MPCDF. The node hosts the full results from all important Virgo simulations (e.g. Millennium XXL, Eagle, Illustris-TNG) and provides web access to the world-wide community via the Millennium database.
Further supercomputers @MPCDF:
Cobra (136960 cores,128 Tesla GPUs)
Raven (128448 cores, 768 Ampere GPUs)
MPA scientists have free access to the MPCDF and are among the top users of the facilities there. MPA and MPCDF are connected via high-speed ethernet and protected by a powerful firewall. The connection to the outside world is via the fast DFN backbone.
MPCDF and MPA coordinate their activities and development plans through regular meetings to ensure continuity in the working environment experienced by the users. The IT-group is made up of four full-time system administrators; users have no administrative privileges nor duties, which allows them to fully concentrate on their scientific work.