An enduring network structure for the PHYSnet data centre at the University of Hamburg

To read a German version of this case study, click here.

With about 39,000 students, the University of Hamburg is one of the largest third-level institutes in Germany. Its 4,760 scientists are supported by about 6,600 employees engaged in building services, libraries, laboratories, medical care and administration. The university’s 150 buildings are distributed across the entire city; however, its centre is located on the Von-Melle-Park campus and in the neighbouring district of Eimsbüttel.

As early as at the beginning of the 1970s, an independent IT system was established in the physics faculty. In the meantime, the PHYSnet data centre as the largest decentralised data centre, which evolved from the original IT system, is at the University of Hamburg and offers IT resources and services to the main body of the physics department. Its range of services include the following: IT coordination throughout the faculty, design and operation of the IT networks, security management of the networks and their components along with the design and operation of centralised departmental, virtualised servers and server services.

Increasing requirements demand a new approach
Until now, the PHYSnet data centre relied on a comprehensive gigabit Ethernet structure in the client environment. In order to meet the increasing performance demands of the clients and cluster systems, the IBM Business Partner pro-com developed a phased concept with low-latency IBM RackSwitch switches for the clients and cluster nodes that is consolidated in a 10-GbE backbone. The backbone has, of course, been designed with redundancy. A particular challenge for the experts at pro-com was to achieve the most cost-efficient conversion of the existing Ethernet infrastructure in the leaf and backbone areas. In view of the limited IT budget in the third-level sector, the decision was taken to opt for a step-by-step conversion of the environment.

pro-com supplies an IBM-based solution that satisfies all requirements
The solution with which the PHYSnet data centre today meets its increased demands is based on the IBM RackSwitch G8000/G8052 switches for server systems connected by gigabit ethernet. These rack switches are small application servers and HPC cluster nodes that provide communication (MPI) across the network and later write the results of the calculation into the storage systems by GPFS. The file servers and GPFS servers, in addition to the IBM switches, are connected by means of a redundant 10-GbE link based on fibre and also on direct copper cables to two IBM 48/64-port 10-GbE switches. This enables the data centre to achieve maximum performance and fail-safety. A decisive advantage for the data centre is the significantly greater performance in the leaf environment for the HPC systems made possible by the low-latency IBM RackSwitch solution. By setting up the redundant 10-GbE backbone, higher performance requirements could also be handled – at no great effort – and a long-lasting structure established.

Result: Greater performance, less complexity
Were pro-com and IBM successful with this project? In the meantime, the PHYSnet data centre can give a positive answer to this question in all respects thanks to the initial experience it has gained. The operator first achieved that the availability and security were significantly increased by the construction of the redundant backbone and by use of the enterprise switch. Yet another plus is the completely redundant (and redundantly configured) structure of the hardware. Furthermore, it was possible to massively increase the overall system performance by converting to 10-GbE technology for the connection of the leaf switches and also through direct support of 10 GbE. Due to the overall consistent management approach toward the systems and the identical fabric operating system, the management of the solution has also been simplified and the complexity for the administration noticeably reduced.