Konrad-Zuse-Zentrum für Informationstechnik Berlin (ZIB)

Story Details

Konrad-Zuse-Zentrum für Informationstechnik Berlin (ZIB) acts as a high-performance data center through the North German Supercomputing Alliance (HLRN-II) and forms part of the D-Grid Initiative, which carries out computationally intensive research projects such as complex scientific calculations for climate research.

The demand for memory and servers at the ZIB is increasing exponentially. The institute's focus is on attaining the fastest possible access speeds. It found that the best solution to this requirement is Sun StorageTek arrays and software.

High-Performance Clusters and Storage Network Replace Legacy Direct Attached Storage Infrastructure

The ZIB has been using Sun servers for some time. Until 2003, these servers were directly attached to storage systems. Then the institute's capacity needs increased sharply due to development of three-dimensional, high-performance visualization applications and saving large amounts of data as an intermediate step in complex calculations. The institute managed just 5 TB of data in 2003, but this rose to 17 TB by 2004 and to 442 TB over the past year. Looking at existing trends, the institute calculated it would have to handle more than 600 TB over the course of 2008.

Carrying On a Tried-and-Tested Partnership

The ZIB replaced its aging direct attached storage infrastructure with a Sun high-availability cluster and a storage area network in 2003. Four years on ZIB opted to expand its storage platform to incorporate four Sun StorageTek 6540 arrays and three StorageTek 6140 arrays. Recalls Wolfgang Pyszkalski, IT services manager at ZIB, other vendors were unable to compete with Sun and LSI when it came to speed. The institute also stipulated that all systems needed to run on a standardized management system. As Sun and LSI Logic were able to deliver the best solution, the ZIB continued to cooperate with its tried-and-tested partners.

Hierarchical storage management (HSM) with StorageTek Storage Archive Manager and corresponding archive or secondary media optimizes the costs and running of the storage infrastructure: The data required most regularly is stored on Fiber Channel disks. Everything else is stored on SATA disks for online access. StorageTek QFS, Sun's multiserver-compatible file system, allows multiple data access simultaneously from different servers, even if they belong to different arrays.

The HSM solution then comes into play. Data that has not been used for a long time is transferred from the cache or the hard drive to the StorageTek SL8500 and StorageTek 9310 PowderHorn tape libraries so the file is still visible to the user but occupies minimal disk space. If a user opens one of these files, the data is automatically recopied from the tape. This saves precious online storage resources.

StorageTek Common Array Manager (CAM) software is part of a centralized management structure for the 6140 and 6540 storage arrays, while StorageTek ACSLS Manager software controls the tape robots. NetBackup from Symantec takes care of the necessary security processes.

Optimized Configuration for Highest Throughput

The required maximum speeds and bandwidths were achieved by intensively studying the system's internal structures and adapting the configuration based on the findings. The most important part of the process was optimizing the distribution of the data on the StorageTek 6540 arrays that are accessed by the HLRN-II users. To this end, each array is connected to the SAN with two Cisco MDS 9134 SAN switches. Two Sun Fire X4600 M2 servers, each with 256 GB memory and eight PCI-E FC ports, are connected to the SAN via these switches.

In addition to providing high availability, the new configuration also guaranteed that there were multiple ways of accessing each query and registered bandwidth speeds of between 900 MB and 3 GB, depending on the size of data transmission block, the performance of the utilized computers and the number of processes running simultaneously.

Effortless Management with Sun StorageTek CAM Management Software

Centralizing its storage environment has made the management process much easier for the ZIB. In the past when errors occurred, the administrator had to be onsite immediately. Now, the StorageTek CAM software records any problems or disk failures while the system automatically activates the available “hot spare” disks in all arrays so that the array can continue to function without interruption. The software then informs the administrators and also displays the location of the fault. The IT staff can now take as much time as they need to swap the faulty disk for a new one without any interruption to service.

After the successful expansion of its storage environment, Pyszkalski is unconcerned about future memory upgrades. When the time comes, the ZIB will in all likelihood once again turn to Sun and LSI.