| HITACHI HOME | UP | SEARCH | HITACHI

rogo HITACHI SR2201 Massively Parallel Processor

-HITACHI SR2201 OFFERS FLEXIBILITY AND HIGH SCALABILITY-

Compact model High-end model

Hitachi's SR2201 series of supercomputers take parallel computing to an entirely new dimension in processing capability and flexible operation. Two model lines are available : Compact and High-end model.
The Hitachi SR2201 High-end model is capable of performing over 600 billion floating-point operations per second (600GFLOPS;theoretical performance) ,and Compact model is capable of performing over 19 billion floating-point operations per second (19GFLOPS;theoretical performance).
The SR2201 uses Hitachi's new high-performance RISC chips,and has a high level of scalability. The SR2201 High-end model ranges from 32 up to a maximum of 2,048 processors,and Compact model ranges from 8 up to a maximum of 64 processors,that are able to meet a wide range of customer requirements.
The SR2201 uses a crossbar switch network arrangement for high-speed message communication,and a pseudo vector processing function that provides a major boost in performance of large-scale scientific calculations.
With its small footprint and flexible power-supply system requirement, The SR2201 compact model can be installed easily in small environments,for example an office or laboratory.
A number of features are included to ensure high reliability and operating efficiency , such as a faulty processor isolation function. A microkernel operating system is used that is based on the Mach 3.0. Transition to parallel processing is facilitated by a flexible program development environment that includes HPF(High Performance Fortran), which enables parallel programs to be written as a natural extension to existing sequential programs , and a matrix calculation line library with parallel processing compatibility.

The SR2201 can be configured with the kind of flexibility that is only possible with a parallel computer, with a choice of processing units ranging from 32 to 2,048 (High-end model),and ranging from 8 to 64 (Compact model) that ensure that all users' computing requirements can be met.



-Features-

1. Hardware
(1) Pseudo-vector processing function
Hitachi's pseudo-vector processing supports high-speed numerical calculations using multiple processing elements.Conventional RISC processors run into difficulty when data for large-scale numerical calculations cannot fit in their caches.The SR2201 series,however ,employs pseudo-vector processing,a pipelining technique which fetches data(operands) directly from main memory (bypassing the cache)and into floating-point registers,without holding up the execution of subsequent instructions.This approach contributes significantly to high performance for large-scale numerical calculations.

sr2

sr3

(2) Three-dimensional crossbar switch network
One of the key architectural considerations in ensuring an excellent cost/performance ratio is the reduction of overhead associated with pre-and postprocessing.The High-end SR2201 line employs an innovative three-dimensional crossbar switch to provide high-speed connection among individual processing elements(PEs).With this switch,there are only three output lines from any PE:one for each of the crossbars.This simple layout achieves almost the same performance as the configuration which interconnects all the processing elements directly ,yet at a much lower cost.
Three-dimensional crossbar switch network
sr4
* Message transmission network for linking PEs.

* A crossbar switch consists of 3 crossbars, one for each axis (X, Y & Z), to create a 2-dimensional or a 3-dimensional structure.

* An X-crossbar switch is capable of switching up to 8 x 8 connections.A Y-crossbar switch and a Z-crossbar switch are capable of switching up to 16 x 16 connections.

* Data transfer rate: 300MB/s in each direction of the dual ports.
(3) Partitioned operation
To ensure flexibility in operation ,the SR2201 series supports partitioned operation.The entire system can be partitioned into a maximum of 8 groups (partitions),each of which can execute its own jobstream independently of the others.
2. Software
(1) HI-UX/MPP UNIX-based operating system
A microkernel operating system is used that is based on the industry-standard Mach 3.0.This strengthens single-system operation, supplements the parallel processing functions of the SR2201 and enhances reliability.
[Single system operation]
Single system operation is possible that provides usability on a par with that of a UNIX workstation with respect to filing systems, process control and network control.
[Scalability]
The number of processors can be expanded from several tens to several thousand, enabling system configuration to be optimized for each task.

sr5

(2) Parallel processing support
"Express" is provided as a support tool for parallel implementation . This includes functions to facilitate programming and performance tuning and the like,such as a message passing library, performance monitor, symbolic debugger, and PVM interface. Additionally,MPI and PVM is provided for parallel programming.MPI provided based on MPI 1.1 standard developed by Message Passing Forum.The forum is a broadly based committee from vendors/implementors including Hitachi and laboratories/universities such as Argonne National Laboratory and Mississippi State University.PVM provided is based on PVM 3.3.10 developed by Oak Ridge National Laboratory,University of Tennessee and etc.Both MPI and PVM perform high performance data communication utilizing SR2201's Remote DMA facility (Direct Memory Access).
(3) Third-Party Software
ISV (Independent Software Vendor) products are supported under REFER (referral of third-party products) projects.

Category

Product name

Supplier

Support status (NOW !)

Parallelization tools,
Debugger
FORGE
Applied Parallel Research,Inc. (USA)
Available
VAST
Pacific-Sierra Research Corp. (USA)
Available
VAMPIR
PALLAS GmbH (Germany)
Available
TotalView
Dolphin Interconnect Solutions,Inc. (USA)
Available
PARMACS
PALLAS GmbH (Germany)
Available
Linda
Scientific Computing Associates,Inc. (USA)
Available
Structural analysis,
Crash analysis
LS-DYNA
Livermore Software Technology Corp. (USA)
Available
MARC
MARC Analysis Research Corp. (USA)
Available
Fluid dynamics
analysis
FLUENT/UNS
Fluent Inc. (USA)
Available
RAMPANT
Fluent Inc. (USA)
Available
STAR-CD
Computational Dynamics Ltd. (UK)
Available
PHOENICS
Concentration Heat and Momentum Ltd. (UK)
Available
STREAM
SOFTWARE CRADLE CO.,LTD. (Japan)
Available (1 node)
SCRYU
SOFTWARE CRADLE CO.,LTD. (Japan)
Available (1 node)
Quantum chemistry
analysis
Gaussian94
Gaussian,Inc. (USA)
Available
AMBER
The University of California at San Francisco. (USA)
Available
MOPAC7.0
Dr.J.P.Stewart (USA)
Available
MOPAC93
Dr.J.P.Stewart (USA)
Available (1 node)
GAMESS-UK
Daresbury Laboratory (UK)
Available
MM2
QCPE (USA)
Available (1 node)
Libraries
IMSL F90
Visual Numerics,Inc. (USA)
Available
IMSL MPI
Visual Numerics,Inc. (USA)
Available (1 node)
NAG F77
The Numerical Algorithms Group Ltd. (UK)
Available (1 node)
NAG F90
The Numerical Algorithms Group Ltd. (UK)
Available (1 node)
NAG Parallel
The Numerical Algorithms Group Ltd. (UK)
Available
Visualization
AVS 5.3
AVS,Inc. (USA)
Available (1 node)
Compilers
KAI C++ TM
Kuck & Associates,Inc. (USA)
Available
@

- Availability -



-Hardware Specifications-
@ Compact models High-end models
System No.of PEs 8 - 64 32 - 2,048
PE-PE network 1,2-dimensional crossbar network 2,3-dimensional crossbar network
PE-PE transfer speed 300MB/s
Total storage(maximum) 64GB 2TB
External interfaces Ethernet,Fast Ethernet,FDDI,HIPPI,ATM
Fast SCSI,Fast/Wide SCSI
PE Theoretical performance 0.3GFLOPS/PE
Addressable memory 64/128/256/512/1024 MB
Cache Primary : 16kB(instruction), 16kB(data)
Secondary :512kB(instruction),512kB(data)
I/O devices Internal hard disk capacity 4.3 GB/drive 2.1 GB/drive
Disk array 4.6 - 66.6 GB/array (RAID5)
Internal DAT capacity 2 - 8 GB(compressed)
External DAT capacity 2 - 8 GB(compressed)
Console display 80 characters x 25 lines
* OSF and OSF/Motif are trademark of the Open Software Foundation,Inc.
* NFS is a trademark of Sun Microsystems,Inc. in the United States.
* Parallel Virtual Machine(PVM) is a public domain package developed by the Oak Ridge National Laboratory in the United States.
PVM is a trademark of the Oak Ridge National Laboratory.
* UNIX is a registerd trademark in the United States snd other countries, licensed exclusively through X/Open Company Limited.
* X Window System is a piece of software developed by the X Consortium,Inc.in the United States.
* Mach is a piece of software developed by Carnegie-Melon University in the United States.
* Ethernet is a trademark of Xerox Corporation.
* Express is a trademark of Parasoft Corporation in the United States.
Design & specifications are subject to change without notice.
To ensure safety and normal operation,be sure to read all instruction manuals carefully before using the equipment.

All Rights Reserved, Copyright (C) 1995,1997, Hitachi, Ltd.
WRITTEN BY General Purpose Computer Division
Author; hpc-ep-info@ml.itg.hitachi.co.jp