What is Atlas?
Atlas is a large computing cluster operated by the Observational Relativity and Cosmology division of the Albert Einstein Insitute located in Hannover, Germany.
It is installed in a 450 m2 basement on the campus of the Leibniz University Hannover, featuring about 1MW of power and cooling capacity and about 4200 rack units of space to mount servers. The external 2x10Gbit/s network connectivity is provided by LUIS, internally, we use a large core switch switch with 768 10Gbit/s ethernet ports connecting most machines at 1 Gbit/s wire-speed via intermediate switches.
Atlas consists mostly of 1000s of compute nodes along about 10 log-in or head nodes. Users can log into these and run scripts and programs which are too large for their local work station or laptop or utilize HTCondor to run their programs on 10,000s of CPU or GPU cores simultaneously.
Data storage is currently handled by about 100 servers which can either host our users’ home file systems or central, “immutable” detector data sets.
Some Numbers
As of 2020, typical metrics for our high-throughput computing cluster currently are
- more than 3,000 individual servers
- more then 80,000 (virtual) compute cores
- more than 20 PByte HDD storage
- 102 water-cooled racks + 8 open racks
- two 800kVA battery-backed UPS and four cooling machines providing stable power and cooling at the 1MW level
A much more detailed page is also available for the curious.
How to get access and start using Atlas?
In general, you need to be either a member of the AEI Hannover data analysis groups or collaborate with us to qualify for using Atlas. If you do not fall into these groups, please contact us by email and let us know why you may need access to Atlas.
Subsequent steps can be found on our dedicated getting started page.