Parsys Limited

Parsys Limited

Status:

unknown.

Overview of Organisation:

Parsys Ltd, Europe's leading Massively Parallel Unix systems manufacturer, is a British-based company formed in 1988 as a spin-off from a joint-European research programme called Supernode. Established within THORN EMI as a separate unit in 1988, PARSYS achieved its first 1 Million pounds sales in 1989 and was launched as an independent company in 1990. It is now 100% owned by Parsys Holdings Ltd, which is in turn owned by Electra Innvotec Limited Partnership, British Gas Ventures Limited Partnership, THORN EMI, and by directors and staff.

Parsys' current range of computers is based on the Inmos T805 Transputer, the SuperNode SN1000 series. This is currently being targeted at the database community in the guise of the ORACLE Parallel Server. The processor technology is due to be upgraded to the T9000, the SN9000 series, which should allow applications to be ported easily with a ten-fold performance increase. They have also recently announced the Concerto system, a joint venture with the other members of the GPMIMD consortium (refer to Concerto under the Meiko section), which is targeted at the high performance scientific market, normally the domain of the vector processors.

Platforms Documented:

Contact Address:

See Also:


SuperNode 1000 series (T800) --- SN1000

Overview of Platform:

The SN 1000 series Architecture --- A large multinode system can be built from a number of nodes in a hierarchical manner via an outer level switching network which is controlled by an Outer Level Controller (OLC). This switching network is used by the nodes to communicate with one another and may be configured by the user. The system is available hosted from a variety of machines: Sun workstation, IBM PS/2, IBM PC AT/XT, DEC microVAX II or VAX. These serve to provide external communications and user access.

Node --- Each node contains two classes of transputer, ``workers'' which perform the computation and ``support'', which control I/O and connectivity. Normally there are 16 workers (though there can be up to 32 in a tandem node) with an equal share of either 512~Kbytes of fast SRAM or 8~Mbytes of DRAM. The support processors are the Disk Server and Memory Server. The Disk Server consists of a control transputer with up to 16 MBytes of DRAM cache allowing disk access and/or tape access via an SCSI interface, 90~MBytes to 1.2~GBytes per disk, 256~MBytes to 2~GBytes per tape. The Memory Server also contains up to 16~MBytes DRAM. This serves the role of a RAM disk local to the node, enabling fast access to areas of common or frequently used data.

The transputer used is the Inmos IMS T805-20 although newer versions of the T8 are expected to be incorporated when they appear.

I/O --- In addition to I/O via the host, each node has a SCSI bus providing access to local disks and/or tape devices. Access is also available via RS232 and X25 interfaces. For high speed access there is a ``Fast I/O Node'' available which enables data transfer at sustained rates of up to 80~Mbytes/second.

Topology --- The OLC system allows the whole machine to be fully reconfigured. In principle this can take place from sub-task to sub-task, enabling optimal communications between processes on processors. The OLC design is such that in a multinode system of less than 64 nodes (1024 transputers) the maximum number of switches through which a message is required to pass is 3. Topologies having up to four-fold connectivity can be formed, e.g. pipelines, broadcast arrays, binary trees, etc.

Operating System --- IDRIS which conforms to the POSIX standard. The Inmos Transputer Development System (TDS) is also available.

Languages --- 3L/Inmos compilers (ANSI and ISO standard) for C, Pascal and FORTRAN. Whitesmith Ltd compilers are also available for C and Pascal. Occam is available.

Performance --- 200 MIPS, 25 MFLOPS to 12000 MIPS, 1.5 GFLOPS on a 64 node system (1152 processors). An SN1370 with eight nodes (each with 16 workers) can sustain 190~MFLOPS.

Data Transfer --- Special I/O nodes can achieve 80~MBytes/second.

Balance --- That of the T800

Scalability --- Currently, the largest SN1000 system consists of 64 nodes.

Compute Hardware:

Interconnect / Communications System:

Memory System:

Benchmarks / Compute and data transfer performance:

Operating System Software and Environment:

Networkability/ I/O System / Integrability / Reliability / Scalability:

Notable Applications / Customers / Market Sectors:

Overall Comments:


SuperNode 9000 series (T9000) --- SN9000

Overview of Platform:

Compute Hardware:

Interconnect / Communications System:

Memory System:

Benchmarks / Compute and data transfer performance:

Operating System Software and Environment:

Networkability/ I/O System / Integrability / Reliability / Scalability:

Notable Applications / Customers / Market Sectors:

Overall Comments:


Concerto (i860)

Overview of Platform:

Compute Hardware:

Interconnect / Communications System:

Memory System:

Benchmarks / Compute and data transfer performance:

Operating System Software and Environment:

Networkability/ I/O System / Integrability / Reliability / Scalability:

Notable Applications / Customers / Market Sectors:

Overall Comments:


hawick@npac.syr.edu
saleh@npac.syr.edu