Path: bga.com!news.sprintlink.net!uunet!concert!ashe.cs.unc.edu!not-for-mail
From: krei...@cs.unc.edu (Amy K. Kreiling)
Newsgroups: triangle.talks,general
Subject: Presentation on IBM's new SP-2 parallel supercomputer - 5/4/94
Followup-To: poster
Date: 25 Apr 1994 14:28:22 -0400
Organization: The University of North Carolina
Lines: 361
Message-ID: <2ph246INN1q8@bodie.cs.unc.edu>
NNTP-Posting-Host: bodie.cs.unc.edu
Keywords: Wednesday, May 4, 10:00am -- 011 Sitterson Hall
Xref: bga.com triangle.talks:111 general:258


From: "Jamie McKee" <jamie...@VNET.IBM.COM>
X-Addr: Office:(919)850-5804 Fax:(919)850-7768
        IBM
        4800 Falls of the Neuse
        Raleigh NC 27609
        email:...@wasvmic1.vnet.ibm.com


Come hear about IBM's parallel supercomputer: The SP-2.

In April, IBM announced the Scalable POWERparallel Systems SP2*, 
the newest entry into its family of RISC-based, parallel information 
systems.  Using processor nodes based on IBM's POWER2* RISC technology.
The SP2 can grow to 128 nodes, with 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.

Two presentations will take place in the Triangle. Tom Haines from IBM's
POWERparallel Development Group will be on hand to present and answer
questions regarding IBM's parallel systems directions and products.

***********************************************************
Wednesday, May 4, 10:00am - 11:30am,
UNC Chapel Hill, Sitterson Hall Auditorium (011)
Contact Amy Kreiling at UNC if you need more detail.
Phone:(919)962-1843, e-mail: krei...@cs.unc.edu

Wednesday, May 4, 2:00pm - 4:00pm,
NC State University, Faculty Senate Room, D.H. Hill Library
Contact Sarah Noell for more details.
Phone: (919)515-5420 email: no...@unity.ncsu.edu
***********************************************************

Agenda:

Overview of POWERparrallel Systems
 - Technology Collaboration
 - Compute intensive market
 - Data intensive market
 - Successes (reference examples)
 - Software Applications

SP2 Hardware Overview
 - Architecture
 - Design Objectives (Palmtops to Teraflops)
 - Features
 - System Overview
 - High Performance Switch

Software Overview
 - System Management
 - Parallel API's
 - System Administration
 - Application Enablers
 - Available Applications (commercial and technical)

Performance
 - LINPACK HPC
 - NAS benchmarks
 - Price/Performance comparisons
 - Future performance enhancements

Presenter:

Tom Haine
2125 Ridgewood Drive
Midland, MI 48642
tha...@detvmic1.vnet.ibm.com (FAX T/L 439-6097)

**************************************************************************
Press Release for the SP-2 POWERparallel System

April 5, 1994

Flexible Processor Configurations, Enhanced Internal Communications,
Terabytes of Data Management

IBM today announced the Scalable POWERparallel Systems SP2*, the newest
entry into its family of RISC-based, parallel information systems.
Using processor nodes based on IBM's POWER2* RISC technology, the new
systems offer customers up to twice the node performance of previous
systems, eight times greater memory and four times greater bandwidth.
The SP2 can grow to 128 nodes, with 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.

POWERparallel System Nodes

The SP2 system can be configured with three different types of processor
nodes.  They include the new POWER2-based "wide" nodes, with seven slots
for I/O and network attachments, one to eight GB of internal disk
storage and 64 megabytes (MB) to two GB of memory per node.

The SP2 has two types of "thin" nodes.  One uses the IBM POWER
Architecture* and the more powerful uses new POWER2-based
microprocessors.  Both thin-node types are identical in local memory,
from 64 to 512 MB, have one to four GB of internal disk storage and four
connectivity slots.  Customers can tailor the system to best suit their
computing needs by choosing a combination of nodes -- a flexibility
never before seen in the high-performance computing market.

The POWERparallel SP2 with POWER2 nodes provides significant performance
improvements over the SP1.  The peak megaFLOPS (equal to millions of
calculations per second) rating of 266 megaFLOPS per node is over two
times greater than the SP1 and testing of technical application codes
has shown up to threefold** performance improvements.

High-Performance Switch

The High-Performance Switch (HPS) and HPS Adapters enable servers on
POWERparallel SP2 nodes to communicate with each other at higher
bandwidth and lower latency.  The new HPS Adapter-2 has a demonstrated
bandwidth of over 30 megabytes/second sustained between every pair of
nodes in the system.  This is more than four times greater than the
performance available with the previous HPS adapters.

In addition, enhanced error detection features that were previously
performed by software have been implemented in the hardware, improving
total system performance by off-loading the processor nodes.  The
latency of the standard communication path has been reduced by a factor
of six.  Actual tests performed with IBM's Parallel Environment Message
Passing Library* (MPL) code have demonstrated that application-to-
application latency is now less than 40 microseconds.

Connectivity

IBM has greatly increased the connectivity options for the POWERparallel
System.  Options such as HIPPI (High-Performance Parallel Interface),
Token Ring, High-Performance Disk Drive Subsystem Adapter, ATM, SCSI-2,
FDDI (Fiber-Distributed Data Interface) and ESCON (Enterprise Systems
Connection) Adapters significantly increase the bandwidth into and out
of the new SP2 system.  Multiples of these high-performance adapters can
be added to the system as connectivity requirements grow.  (See details
below.)

This dramatic acceleration in communications technology, such as HIPPI,
allows the POWERparallel system to access, process, analyze and store
vast amounts of all kinds of information, including digital libraries
video.  Connectivity between the POWERparallel system and IBM
System/390*, via ESCON and Block Multiplexer Channel Adapters, provides
customers with increased flexibility in determining the best and most
cost-effective allocation of computing resources.

IBM currently intends to announce and make available additional options,
such as Fibre Channel, SCSI Fast/Wide Adapter and ESCON Channel Emulator
adapters during the second half of 1994.

IBM Storage Options

The flexibility of the new models of the POWERparallel system SP2 offers
commercial customers scalable information processing solutions, as well
as providing scientific/technical customers with the data-intensive
capabilities they need, at excellent price/performance levels.

The following storage devices from IBM's Storage System Division are an
important part of these solutions with their high-availability, high-
performance data storage capabilities.  They include:  the IBM 9333
High-Performance Disk Drive Subsystem, which provides disk storage
using a serial link architecture, the IBM 7135 RAIDiant Array, a fault-
tolerant RAID solution; the IBM 9570 High-Performance, High-Capacity
Disk Array Subsystem, which provides high-performance peak data rates
of 60 MB/second; the IBM 3995 Optical Library Dataserver, for access to
information that has traditionally been stored on paper, microfiche,
tape or diskette; and the IBM 3494 Tape Library Dataserver, which
supports multiple computing platforms with up to 7.2 terabytes
(compacted) of storage.  Each of these devices is described in detail
below.

IBM Connectivity Options

ESCON Control Unit Adapter - provides the ability to attach the SP2
nodes to IBM ESCON channels of the System/390 mainframe.  The ESCON
Control Unit attaches directly to an ESCON channel, providing fiber
optical links using LED technology, and also attaches to ESCON
Directors (fiber optic switches) to allow for a large number of
connections.  Host operating systems supported include:  VM/ESA,
MVS/ESA and AIX/ESA.

Token-Ring High-Performance Network Adapter - designed to allow an SP2
node to attach to 4-megabytes/second (Mbps) or 16-Mbps Token-Ring local
area networks.  This Token-Ring adapter is cable-and-network compatible
with all IBM PS/2 Token-Ring adapters.

HIPPI (wide nodes only) - enables high-speed connectivity for the SP2 to
supercomputers, IBM RISC System/6000s, HIPPI fiber optic extenders,
HIPPI switches, IBM 9570 Disk Array and other disk arrays and tape
subsystems.

The HIPPI feature provides an efficient simplex/duplex point-to-point
HIPPI interface achieving peak rates of 800 Mbits/second
(simultaneously in each direction).  The adapter can be used for either
communication or storage-channel applications.

TURBOWAYS 100 ATM Adapter - enables TCP/IP applications to work in an
ATM environment.  The ATM adapter is a Type 5 Microchannel adapter for
the RISC System/6000 or SP2.  The initial release supports AAL-5
adaptation layer interface.

SCSI-2 Differential High-Performance External I/O Controller - provides
attachment of external SCSI-2 differential devices.  It supports up to
a 10 MB/second data rate for SCSI-2 differential devices and supports
Command-Tagged Queuing.

FDDI - Fibre Dual-Ring Upgrade - will allow customers to attach to both
the primary and secondary FDDI ring.  FDDI has a maximum transfer rate
of 100 Mbps.

FDDI Single Ring Single Attach Station (SAS) Adapter - attaches the SP2
directly to a primary ring of a FDDI network via a FDDI concentrator.
The FDDI concentrator offers additional protection by isolating the
network from routine on/off activity and individual failure of the SP2
processor nodes.  FDDI has maximum transfer rate of 100 Mbps.

Block Multiplexer Channel Adapter - allows the SP2 wide nodes to
communicate with a S/370 or S/390 host using the Block Multiplexer
Channel Adapter cable.  TCP/IP is supported and the IBM TCP/IP
Version 2.2 for MVS or VM is required at the host, if running MVS or VM.
This connectivity allows the SP2 to be used as a gateway between the
S/370 or S/390 host and the downstream networks that consist of a LAN or
WAN.  Therefore, distributed systems can access the MVS or VM resource,
Network File System (NFS) and the TCP/IP applications resident at the
System/370 or System/390 hosts.

Ethernet High-Performance LAN Adapter - a high-performance Micro Channel
architecture Bus Master adapter to attach an SP2 node to Ethernet
networks.  This adapter is designed to provide connection to a 10-Mbit
Sense Multiple Access/Collision Detection (CSMA/CD) Ethernet network
and is compatible with IEEE 802.3 and Ethernet external interfaces.
This adapter is standard on all SP2 wide nodes.

High-Performance Disk Drive Subsystem Adapter - allows attachment of
four (per adapter) IBM 9333 High-Performance Disk Drive Subsystems to a
POWERparallel system.  Maximum storage per subsystem is 8 GB, increasing
the maximum storage per High-Performance Disk Drive Subsystem Adapter to
32 GB.  Maximum transfer rates are either 40 MB/second or 80 MB/second.

IBM Storage Devices

IBM 9333 High-Performance Disk Drive Subsystem - provides high-
performance disk storage using a serial link architecture.  This
architecture delivers very high sustained I/O rates by mixing data and
high sustained I/O rates by mixing data and commands in full-duplex
mode, thereby eliminating bus contention and maintaining high
performance in high-capacity systems.

The 9333 comes with two 2.0-GB Serial-Link Disk Drives and can be
expanded to 8 GB with two additional disk drives.  Up to four 9333
subsystems with a total storage capacity of 32 GB can be attached to a
single High-Performance Disk Drive Subsystem Adapter.  The adapter is
designed to attach to four 9333 subsystems without performance
degradation.

IBM 7135 RAIDiant Array - provides fault-tolerant disk storage with
advanced features such as hot-pluggable disk drive modules and
concurrent maintenance of disk drives, controllers, power supplies and
cooling fans without system disruption.  All configurations of the
RAIDiant Array Model 110 implement multiple, concurrent RAID levels
(0, 1, 3 and 5).  An optional standby controller configuration provides
automatic takeover by the second controller if an error is detected in
the active controller.

The Model 110 includes 2.6 GB of disk storage, which can be expanded to
a maximum capacity of 60 GB.  For easy maintenance and expansion,
individual disk drive modules can be replaced by the customer.

IBM 9570 High-Performance, High-Capacity Disk Array Subsystem - a high-
performance, high-capacity distributed-storage RAID product.  SP2
attachment is via the industry-standard HIPPI interface with IPI-3
protocol for high-performance peak data rates of 60 MB/second or more.
It has programmable support for both RAID 1 and 5 operation on the same
facility and is rack-mounted with a storage capacity from 13 GB to
232 GB.  Increasing the capacity does not impact floor space.

IBM 3995 Optical Library Dataserver - provides cost-effective, direct
access to information that has traditionally been stored on paper,
microfiche, tape or diskette.  Both rewriteable and permanent Write
Once, Read Many (WORM) media are supported within the same library
using IBM's industry-leading 1.3-GB drives.  Capacities range from 20 GB
to 188 GB.

IBM 3494 Tape Library Dataserver - provides fast and dependable disk-to-
tape backup operations and large data-archive and recall facilities.
The IBM 3494 Tape Library Dataserver configuration can grow to 7.2
terabytes (compacted) while supporting multiple computing platforms,
including AIX/ESA, AIX/6000, OS/400, MVS/ESA, VM/ESA and VSE/ESA
simultaneously.

World-Class Systems

IBM's POWER Parallel Systems business unit produces world-class
scalable, parallel information systems for commercial and
scientific/technical customers.

The IBM Scalable POWERparallel Systems 9076 SP1 and SP2* feature design
and performance leadership, offer exceptional reliability and
versatility, and deliver high-performance computing at workstation
price/performance levels.

Headquartered in Somers, N.Y., IBM's POWER Parallel Systems business
unit also draws on resources from the IBM Large Scale Computing
Division, IBM RISC System/6000 Division and IBM Research.


IBM TO COMPLETE POWERPARALLEL RISC-BASED SOLUTIONS WITH AWARD-WINNING
DB2 DATABASE MANAGEMENT SOFTWARE

April 7, 1994

IBM has announced that its DATABASE2 AIX/6000* (DB2/6000*) will support
the company's POWERparallel RISC-based computers.  The DB2/6000 Parallel
Edition will fully exploit the parallelism of IBM's hardware,
significantly reducing response time for complex queries and
transactions.  It will support very large databases, hundreds of
gigabytes, across multiple nodes of a POWERparallel system.

"The DB2/6000 Parallel Edition is the first to recognize near linear
scalability on a 64-node processor," said Janet Perna, division director
of database technology, IBM Software Solutions Division.

"This allows customers to grow their applications and processing power
incrementally without disrupting their existing systems."

IBM disclosed its DB2/6000 plans as the company's Large Scale Computing
Division and POWER Parallel Systems business unit announced the next
generation of parallel processors and software, including enhancements
to the Scalable POWERparallel Systems SP1* and SP2*.

"We're very pleased to see that the DB2/6000 Parallel Edition will fully
exploit the scalability of the POWERparallel platform to provide an
industry-leading, high-performance database system," said Irving
Wladawsky-Berger, general manager, POWER Parallel Systems.

"I'm convinced that the combination of parallel DB2/6000 on the SP
platform will create new and wonderful applications for customers,
allowing them to analyze and manage vast amounts of all kinds of data."

The DB2/6000 Parallel Edition will allow customers to grow databases by
adding nodes incrementally and with near linear performance
improvements.  Availability will improve because failures can be
isolated to an individual processor.  While the failed node is
recovered, the rest of the system can continue to operate.

To exploit multi-node parallelism in the complex query market, the
DB2/6000 Parallel Edition will decompose SQL queries into many parts,
which then will be run on multiple nodes in parallel.  This can reduce
response time significantly.  In addition to queries, all functions can
be performed in parallel, such as updates, index creation, backup and
restore.

Customers' investments are protected because parallel exploitation will
be performed automatically without costly or time-consuming changes to
existing applications.

IBM recently demonstrated the DB2/6000 Parallel Edition at several
industry trade shows, including UniForum in San Francisco and CeBIT in
Europe.  IBM DB2/6000 Parallel Edition is currently in beta testing with
a selected set of customers.  Availability will be announced later this
year.