Created in 1988 as part of a national initiative in high-performance computing, The Centre for Development of Advanced Computing, headquartered in Pune, India – Poona to gnarled India hands, and the town where ICL Plc’s Indian affiliate has its plant – has launched Param 9000, which it calls an OpenFrame flexible supercomputer architecture. The OpenFrame architecture enables the integration of homogeneous and heterogeneous nodes and unifies cluster computing with massively parallel processing, it claims. The centre says that because of the inherent property of its OpenFrame Architecture, upgrades with either new processors or advances in interconnect technology – can be effected with minimal difficulty. The nucleus of OpenFrame is a scalable interconnect that can support 1,000 nodes or more. The interconnect is a low latency, high bandwidth point-to-point link. OpenFrame has a variety of input-output and networking interfaces, including Ethernet and fast and wide SCS interface. It uses an SBus-to-VME bus interface. The Param 9000/SS uses SuperSparc nodes and is the first OpenFrame offering. The software environment, Paras 9000, is the centre’s program development environment, and includes parallel programming extensions to a Solaris 2.X operating system. It has optimised a Mach microkernel, and offers a parallel high performance file system, standard and enhanced compiler optimisations and parallel libraries.
Bangalore and New Delhi
The operating system configures the system into service, compute and input-output partitions. The user logs onto a service node and uses the Paras program development tools to develop a parallel application. The Paras 9000/SS supports two main massively parallel processing programming models: data parallelism and multiprocess parallelism. The service nodes run Solaris, while the Paras microkernel is replicated on the compute nodes and includes Mach-like process management, enhanced exception handling, a virtual memory model and interprocess communications. It supports threads, ports and port groups and virtual memory regions and can be accessed by a variety of hosts. The interconnect uses a packet switching wormhole router. Each switch can establish 32 simultaneous non-blocking connections to provide a sustainable bandwidth of 320Mbps. The communication links are Peer-to-Peer Protocol. Current compute nodes use 60MHz SuperSparc IIs with 1Mb external cache, 16Mb to 128Mb memory, one to four comm unication links and related input-output devices. With some 40-odd applications in the making, the centre says it will support parallel database management, complex query decision support system and video on demand in the near future. Parallel libraries, application kernels and benchmarking, and parallelisation and conversion of standard packages are under way. The centre has 250 employees in Pune, Bangalore and New Delhi, and is seeking partnerships worldwide. A Param 9000/AA series using Digital Equipment Corpo’s Alpha RISC is in the works, as is Param 9000/PP, with PowerPC nodes. The Param 9000/SS is shipping now; no prices.
