IBM Announces New Family of POWERparallel Systems

Parallel RISC-Based Computers Drive New Market Segment

April 5, 1994

IBM today announced the next generation of its RISC-based high performance parallel computer, the Scalable POWERparallel Systems SP2*. The SP2 represents a new breed of information system which combines the numeric-intensive processing capabilities of scientific and technical computers with the storage and analysis strengths of commercial systems.

Based on IBM RISC System/6000* microprocessor technology and running AIX/6000*, IBM's UNIX** operating system, the POWERparallel System SP2 offers customers unmatched system flexibility, with options to scale, or grow, each component of the system as needed -- from the processing nodes to the High Performance Switch [ http://ibm.tc.cornell.edu/ibm/pps/switch/index.html ], which enables parallelism. Nearly all of the 10,000 available AIX applications can run on the POWERparallel system, as well as over 65 new commercial and technical parallel applications.

Blending the best attributes of workstations and high performance computers, the new POWERparallel systems scale from four to 128 nodes, and offer significant flexibility in configuring the systems by combining nodes -- some optimized for storage and communications and others optimized for price/performance -- within a frame. The systems also offer POWER Architecture* and POWER2* [ http://ibm.tc.cornell.edu/ibm/pps/power/power2/index.html ] processor choices, immense storage capacity and increased network and disk connectivity options.

The new POWERparallel systems POWER2 processors will offer customers twice the processing power of the current system, eight times greater memory and four times greater bandwidth (transmission capacity). A 128-node system has a peak performance of 34 gigaFLOPS (equal to billions of calculations per second), 256 gigabytes (GB) of internal memory and 1,024 GB of internal disk storage.

Wide Nodes, Thin Nodes

Introducing the concept of "wide" and "thin" nodes, the new systems will provide both commercial and scientific/technical customers with exceptional system flexibility.

The new SP2 models are based on either 62.3 megahertz (MHz) POWER Architecture or 66 MHz POWER2 microprocessors and are offered in both wide and thin node implementations -- both of which have exceptional computing capabilities. The POWER2 microprocessors will offer customers 266 megaFLOPS of processing power (equal to millions of calculations per second) -- over twice the processing power of the current system processors.

Wide nodes offer extensive storage capacity and communications capabilities and are based on POWER2 microprocessors. They have seven slots for I/O and network attachments for a broader choice in connectivity options, up to eight GB of internal disk storage and up to two GB of memory per node, providing very high compute capacity, greater memory and wider bandwidth.

These new systems are well-suited for database query, decision support, business management and batch business applications. They can also offer the highest compute performance, making them ideal for scientific/technical customers who want to scale to massively parallel computing performance -- over 100 gigaFLOPS.

The SP2 has two types of thin nodes. One uses the IBM POWER Architecture and the more powerful node uses new POWER2-based microprocessors. Both thin node types are identical in local memory, from 64 to 512 megabytes (MB), have one to four GB of internal disk storage and four connectivity slots. Additionally, the number of processors can be increased incrementally to meet computing power requirements for interactive, batch, serial and parallel jobs, which can all be run simultaneously.

Demonstrating the flexibility of the POWERparallel family to expand with new technology, the new systems can co-exist with SP1 systems, or current SP1 systems can be easily field upgraded.

To continue the success of the POWERparallel System, IBM is investing in several key technology areas including: job management software; systems management software; System/390 co-existence; parallel application development with independent software vendors; and, the High Performance Switch (HPS) and HPS adapters -- now offering greater bandwidth and lower latency (command response time).

Many of the advancements offered with the SP2 are the results of rigorous testing and evaluation of the SP1 by scientific and technical customers, whose advice is reflected in SP2 technology and software.

Nearly Unlimited Connectivity

Customers will be able to connect to nearly every network they may have installed, empowering the POWERparallel system as a full-function server for client/server applications. Offering broader connectivity and higher communication speeds, newly available Micro Channel adapters will include: ESCON*, Token Ring, HIPPI (High Performance Parallel Interface), High Performance Disk Drive Subsystem Adapter, ATM, FDDI (Fibre Distributed Data Interface) and SCSI-2. IBM also announced its intention to support the ESCON Channel Emulator, SCSI-2 Fast/Wide adapters and Fibre Channel adapters.

IBM's High Performance Switch (HPS) and HPS Adapters will enable the new POWERparallel processing nodes to communicate with each other at an exceptionally higher bandwidth and lower latency. The new systems have demonstrated greater than 30 megabytes/second bandwidth and less than 40 microsecond latency.

Storage Systems

A wide range of storage devices from IBM's Storage Systems Division offers enormous storage capacity, high performance and high data availability for POWERparallel systems. Included in the offerings are: the 7135 RAIDiant Array; the 9570 High Performance, High Capacity Disk Array Subsystem; the 3995 Optical Library Dataserver; and, the 3494 Tape Library Dataserver. Up to 31 terabytes (trillions of bytes) of disk storage can be attached to the new SP2 system.

Improved System Software, New Applications

With upgrades in system software for the current and new systems, customers will benefit from new functionality and enhanced system management, full versions of AIX/6000 running on each node, an integrated parallel development environment, full parallel job scheduling, a parallel file system and a set of tools to support parallel database implementations.

New IBM applications include: IBM DB2/6000* and DB2/6000 Parallel Server*; IBM CICS/6000; IBM Encina (based on Transarc technology); IBM NetView/6000*, a network management tool; IBM Visual Data Explorer*, for 2D and 3D visualization of complex data; and IBM Optimization Subroutine Library (OSL) Parallel Support*, software for determining an optimal allocation of limited resources.

Nineteen new independent software vendors have indicated that they intend to port their applications to run in parallel or parallel cluster mode on the POWERparallel system, bringing the number of software vendors supporting the system to over 65.

Additional parallel software for the SP2 announced today includes ORACLE7.1 Parallel Database software from Oracle Corporation; SYBASE Navigation Server from Sybase, Inc.; Tuxedo from IMC Corporation; and SAP R/3, a client/server-based package of business applications software from SAP AG.

Price/Performance

The POWERparallel SP1 has the best price/performance in the industry based on the Numerical Aerodynamic Simulation (NAS) benchmark performance measured per million dollars. The POWERparallel SP1 has a performance rating of 27.60 on the EP (Embarrassingly Parallel) kernel benchmark, with its closest competitor, the Cray T3D, rating at 8.52.***

The POWERparallel system also posted the best price/performance on the NAS SP pseudo-application benchmark. Other systems rated include the Thinking Machines CM-5, the Intel Paragon and Kendall Square Research's KSR1.****

Performance numbers for the new SP2 system, which are expected to be significantly higher, will be released when the system becomes generally available.

The first SP2 systems will be shipped later this month with volume shipments planned for this summer.

Entry into Commercial Arena

Many businesses today have vast quantities of stored data as a result of their day-to-day business operations, including sales history and inventory data, and customer, supplier and product information. The new POWERparallel system will analyze this information to identify marketing trends and buying patterns, enabling companies to more effectively design marketing strategies.

Commercial customers will use the new system for a number of strategic functions including executive decision support; LAN consolidation; centralized management of distributed systems; quick-response "data mining" applications in financial investment research, banking, insurance and retail industries; complex business simulations; "just-in-time" warehousing; extensive digital libraries; and, massive multimedia and video-on-demand applications.

An increasingly popular use of the POWERparallel system for strategic data mining applications is fraud analysis, targeted at the insurance, banking and credit card industries, which analyzes huge databases and can detect possible fraud through pattern recognition.

Widely Popular Among Scientific & Technical Customers

The SP1 has enjoyed wide acceptance in the scientific/ technical environment. The first two SP2 systems will ship to the Cornell Theory Center [ http://www.tc.cornell.edu/ ] and the Maui High Performance Computing Center [ http://www.mhpcc.edu/mhpcc.html ].

The Cornell Theory Center, a National Science Foundation-funded supercomputing center at Cornell University, will be the first customer to receive the SP2.

The Maui High Performance Computing Center SP2 system which is being shipped in April.

Scientific and technical customers use the POWERparallel system for such applications as seismic analysis, molecular modeling of drugs, computational fluid dynamics, plasma physics and other Grand Challenge problems.

As with the SP1, scientific and technical users will continue to push the envelope of performance, with the Cornell Theory Center and the Maui High Performance Computing Center installing systems with over 100 gigaFLOPS peak performance by the end of the year. IBM works closely with its partners to learn how the system reacts in such demanding environments and applies that knowledge toward new generations of systems.

IBM's Large Systems Strategy

The POWERparallel system plays a major role in IBM's large systems strategy where it is one of two available platforms. The POWERparallel system co-exists -- sharing workloads and data -- with IBM's System/390 and two MVS-based S/390* parallel systems, the S/390 Transaction Server* and the S/390 Query Parallel Server*.

IBM POWERparallel systems are the high end of the IBM RISC System/6000 product line, and thus represent the high end of IBM's "palmtop to teraFLOPS" strategy.

* Trademark or registered trademark of International Business Machines Corporation.

** UNIX is a registered trademark of X/Open Company Limited.

Other product names may be trademarks of their respective companies.

*** Competitors' price and performance data are from David H. Bailey, Eric Barszcz, Leonardo Dagum and Horst D. Simon, "NAS Parallel Benchmark Results 10-93, RNR Technical Report RNR-93-016, October 27, 1993" and update of February 28, 1994.

**** Performance measurements for the IBM POWERparallel system are the result of tests done in a laboratory environment on a production level (generally available) machine at IBM Kingston using Message Passing Library (MPL)/p. While these values should be indicative of machine performance, no warranties or guarantees are stated or implied by IBM. These measurements are offered only as an indicator of performance.

The NAS benchmark suite is a commonly used indicator of parallel computer performance that has been developed through the Numerical Aerodynamic Simulation (NAS) program at NASA's Ames Research Center. The benchmark suite is comprised of five parallel kernel benchmarks and three simulated computational fluid dynamics application benchmarks.

Competitors' performance data is from David H. Bailey, Eric Barszcz, Leonardo Dagum and Horst D. Simon, "NAS Parallel Benchmark Results 10-93, RNR Technical Report RNR-93-016, October 27, 1993" and update of February 28, 1994.