|Exhibition and Industrial Session: Siemens-Nixdorf AG|
The Reliant RM1000 is SNI's massively parallel processing high end system within its RM series. It is developed by SNI's affiliated corporation Pyramid Technologies and especially suited for applications like Decision Support Systems and Data Warehousing, which are based on big distributed relational databases, e.g. Oracle or Informix.
The system provides a wide range of scalability: Hundreds of compute nodes and terabytes of disk space are configurable. It is driven by UNIX System V.4 according to XPG/4 instructions. The system's backbone is a redundant 2-dimensional network called MESH. The compute nodes reside on the intersection points. MESH provides a high bandwidth of 4 x 40 MB/s effective.
The compute nodes are currently based on the MIPS R4400 processor (150 MHz) and are equipped with 256 MB RAM. They are arranged in cells; at most 6 nodes fit into a cell. Furthermore a cell can hold up to 24 disks with 2 or 4 GB capacity each. The disks are connected to the compute nodes via dual-hosted SCSI busses.
Messages on the MESH network are forwarded according to the principle of "wormhole routing", which is a simple and deadlock-free process. A single network appearance cares for fault resilience. A system-global device name space ensures that each compute node can access each mass storage device. A virtual Ethernet concept guarantees load balancing between the compute nodes. The scalability becomes clear by the conclusion "More CPU's - more MESH connections - more parallel transfers". Scalability reaches far, but is limited: currently there are no problems with up to 200 CPUs.
The system administration is centralized on one console workstation. It is carried out on one GUI-based resource and user management.
Further development steps will include 200 MHz CPU's equipped with 512 MB RAM.