History

IW People - History

Although the history of Computer Systems at IW has now come to an end, it was a long and productive one. This page provides a brief summary of the projects that we worked on. Many thanks to Bob Hanson, who compiled the bulk of this information and is responsible for any material not otherwise credited. Specific contributors are identified by their initials:

  • (-JCB) James Blasius
  • (-BG) Brad Grande
  • (-MG) Mike Grinn
  • (-RJH) Bob Hanson
  • (-RS) Rich Strebendt
  • (-SJC) Shu-Jen Chen

If you would like to submit a more detailed description of any of these projects, just email it to me and I'll paste it in.


The many Names of IW

  • Bell Labs Processor Division
  • Bell Labs Computer Division (-RS)
  • AT&T Information Systems
  • AT&T Data Systems Group
  • AT&T Computer Systems
  • National Cash Register/the NCR Corporation
  • AT&T Global Information Solutions

Opened November 25, 1980
Closed December 15, 1995

The many Presidents of IW (-MG)

I was among the first wave of folks to move into IW from IH (they didn't have to call it "main" then, 'cause the complex only had one building!) back in December of 1980.

As a hobby, I used to keep a list of the many "presidents/heads/chairmen" of the AT&T computer effort as they came and went (pretty regularly, too). As you can see, we went through 12 of them in 10 years. Anyway, without any guarantees of perfect historical accuracy on the dates, here is my list of AT&T computer presidents:

Computer Systems Presidents
1980-85 Jack Scanlon Bell Labs, Processor Division
1986 Archie McGill American Bell, ABI, the precursor to AT&T-IS
1987 James Edwards AT&T-IS
1988 Vittorio Cassoni may he rest in peace
1989 Bob Kavner AT&T Computer Systems?
1990 Rich McGinn / Gordon Bridge "McBridge" to their friends
1991 Chuck Exley NCR (still laughing, somewhere in the Caribbean)
1992 Gill Williamson NCR
1992-93 Jerre Stead AT&T-GIS
1994 V.A.Cant Bill O'shea acting head
1995 Lars Nyberg NCR again... Lars a.k.a. "the count"
1995 end of the road for good ol' computer systems

The many Projects of IW

3B20D - DMERT

The world's most reliable computer: everything is duplicated, and the active-standby switchover can be activated in mid-CPU instruction. Still used today as the heart of the 4ESS, 5ESS, and Autoplex systems, it is the only computer to withstand a 7 Richter earthquake in both Lawrence Livermore Laboratory tests, and actual field tremors in southern California.

LAMP

A system (for internal use) of hardware verification and fault detection test verification software. It was originally chartered to do full fault simulation of the 1A processor, but was designed to simulate any circuit of <64K gates of digital logic. As the needs of hardware developers evolved, the system evolved to include a True Value (design verification) simulator, a simulator of shorted-paths faults, as well as an Automatic Test Generation tool. At one time, the LAMP system was the most heavily used program on the TSS systems. The circuit description language invented for LAMP (the Logic Circuit Analyzer Language - LOCAL) was standardized for all Bell Labs-developed simulators under the name Logic Simulation Language (LSL) and is still in use on other simulators nearly 30 years after the LOCAL compiler was first run. LAMP was started in 1969 and continued in use into the 1990's, both as LAMP and its later version called LAMP-II. (-RS)

AIDE

An internal software product for simulating interconnected events.

3B20S - UNIX

The normal commercial availability version of the 3B20D. Like the VAX 11/780, a standard by which computers of that era were judged (1 MIPS CPU). It saw duty in telecommunication network applications which did not have to be fault tolerant. (-RJH)

The 3B20S is not a highly available 3B20D but it has all the fault detection mechanisms. When a hardware fault is detected, a 3B20D will switch to the stand-by side in seconds; the 3B20S has no place to switch to but will stop the processor from screwing up. At that time, many so call fault tolerant computer in the world did just that. (-SJC)

TNET

An attempt to build an X.25 message switcher using a network of special purpose processors connected by a coaxial cable bus. The bus was designed to run at 32Mbps. This project was killed (replaced by the CNI ring) but that work became the basis for the 3BNET product (after the bus was modified to be Ethernet-compatible). (-RS)

3BNET

The Bell Labs version of the Ethernet concept, with several improvements. Embraced ISO OSI protocol stack and provided basic networking between machines over thick co-axial cable (10 Mbps).

3B5

The world's first minicomputer based upon the world's first 32-bit microprocessor, the BELLMAC 32A. It featured removable CDC Lark hard disk platters years ahead of Bernoulli cartridges, 8" form factor peripherals instead of 14" peripherals, high-density semiconductor RAM, and a modern peer-oriented system bus which could accommodate multiple CPUs. It had a high-performance port of UNIX SVR2 and was used in telecommunication billing applications.

3B40/3BX

A loosely coupled system made up of Bellmac32 processors in a 3B20-like architecture, with a goal of 5 times the 3B20S in performance. A separate effort from ALICE (see below), it was canceled due to architecture problems.

ALICE

An asymmetric multiprocessor project using the 3B5, with 3 CPUs and an experimental UNIX kernel with the process scheduler adjusted for multiprocessing. It was way ahead of its time, and demand paging took development priority, so the software work was never finished.

3B15

The UNIX demand paging version of the above 3B5, with upgrades in componentry. The 3B15 had a long life in telecommunication applications such as in Denver's System 75 and 85 PBXes, and continued in billing and OA&M applications.

FATE & QUARTZ

The Fialkoff Asynchronous Terminal Emulator, and its successor QUARTZ, were terminal emulator software products which drove user commands into a system under test over either individual RS-232 lines or TCP/IP Ethernet network sessions. Many benchmark efforts such as CI-20, QUARTZ-Gaede, and TPC Benchmark A used QUARTZ, which became commercially available in its later life (then sold to East Coast Computer Systems). In the last large-scale benchmark test, up to 19,000 users on multiple driver systems were driven. QUARTZ is still used around the IH complex today for test load generation.

BUSTER Test Management System software

Another internal tool which eventually became commercially available, BUSTER manages system test administration. It produces test execution reports, rolls up test statistics, and maintains a database of results.

StarLAN

A pioneering low-cost premise data network product (similar to Ethernet) which used ordinary twisted pair wires for 1 Mbps operation. Cheap to install and maintain, StarLAN was the predecessor of later IEEE standard thin-cable and twisted-pair products. The latest version (StarLAN-10) was compliant to IEEE 10 Base T standards (10 Mbps) and could still operate over twisted pair from a fiber-optic hub.

StarGROUP/RFS

The Bell Labs version of the ISO OSI protocol stack. StarGROUP encompassed all the protocols up through level 4 (TLI) and applications on top of these such as Email and remote messaging. The Remote File Sharing facility was the UNIX System V method for remotely mounting files across the network, much like NFS, but more tightly integrated into the UNIX fabric. All lower level protocol work on StarGROUP was done at IW, and RFS was done in NJ and tested here.

3B2

One of the most daring projects yet. All new in concept, featuring the latest in compact high-performance hardware technology, this machine set the standard for years in a new class of systems later called supermicrocomputers - traditional minicomputer power shrunk to desktop size. It became the best-seller in its market niche (midrange UNIX machines in communication-oriented applications) and had a life cycle of about 8 years. New technology included asymmetric multiprocessing, a virtual-address-based cache memory, all SCSI-based peripherals, and high-performance communication boards each featuring their own CPU for low-level I/O processing. Later models also had magneto-optic platter storage (5.25" form factor). The UNIX kernel was SVR3 and adjusted for multiprocessing - first just the scheduler (3B2/600) and later even for I/O (3B2/1000). Later models had cool codenames like Falcon (3B2/600) and Galactica (3B2/1000). It landed the world's biggest potential contract (quoted at $2 Billion), but of course, the Air Force never bought that many. The last version of the system, codenamed Phoenix, contained a MIPS CPU because WE 32000s had been discontinued by AT&T Microelectronics.

XL/CRISP

Murray Hill had developed the C Reduced Instruction Set Processor (CRISP) as the latest in RISC technology. XL was an IW project to implement the instruction set architecture in ECL (emitter coupled logic). ECL is extremely fast, extremely hot (temperature), and the least reliable of digital circuit technologies. Mainframes use ECL and require liquid cooling plants attached to the processors. Breakthroughs in reduction of CMOS geometries and gains in clock speeds possible on-chip made the new VLSI chip a somewhat slower but whole lot more reliable and cheaper alternative to implementation. XL was canceled due to ECL technology problems. A CRISP chip was produced later (in Apache timeframe) and tested in a prototype board by an IW person, but production was never completed due to Microelectronics leaving the CPU business. The Murray Hill researcher mainly responsible for the CRISP went to another company, and a third company patterned its RISC chip off "white papers" delivered at conferences by this researcher.

Intercept

The bus and memory of XL were saved and used on the Intercept systems. Intercept 25 was 1-5 processors made out of Bellmac32-B (WE 32100) processors. Intercept 27 was the High Availability version. IW made a few prototypes before the project was canceled in favor of Apache (see below) and because product management didn't know how to manage or market it.

COMPANION

The Bell Labs attempt at an early laptop computer. Large by today's standards, it was nevertheless a full UNIX system featuring the WE 32000 32-bit CPU. It was never put into production because there were no applications ported to it and no one wanted a multi-user O/S like UNIX on a single-user carry-around box.

ALEXANDER

The daring 3B2 reduced to less-than laptop size! Featuring a 14 MHz WE 32100, the latest ASICs, all 3.5" peripherals, and a unique stackable opto-coupled internal bus, traditional minicomputer power was reduced to a box of only 1 square foot and 5 inches high! It also featured EPROM modules as a storage medium (just like modern video gamepacks), with built-in compression and decompression code to increase capacity. It was so small, so unique, and so outrageous, it had no market, other than internal sales to the IH comp center for a troff engine application and a few other internal customers.

FOCAS

French's Optimized Computer Analysis System - an internal tools project which turned a 3B2 into a programmable logic analyzer. Custom hardware and software enabled performance analysis of any other system by attaching the probes and programming data collection.

SCSI Mud Turtle

Another internal tools project which provided complete analysis of SCSI Bus activity. Mud Turtles were used for years in all other projects which had SCSI peripherals.

The Native American named products - Apache, Blackfoot, Cherokee, Dakota, and Illini.

These were specifications for large multiprocessors and network products to go with them. The projects were started over time in roughly alphabetical order. Of these, only Apache was ever finished. The Illini concept later morphed into the Yellow machine (see below).

3B4000 Apache

An attempt at large-scale multiprocessing not using a shared-memory paradigm, the Apache project produced the biggest machine in IW history (at least with respect to hardware). Unique technical breakthroughs like single-virtual machine image at the system-call level and a distributed UNIX kernel were overcome by market forces when many third party suppliers (mostly database vendors) refused to port to its new architecture, and by internal squabbles between product management and development. Called a "network in a box", it found specialized applications, and was part of the big Hughes FAA bid to overhaul the nation's air traffic control system (which IBM won and still hasn't delivered on). (-RJH)

It survived a long time after the 3B4000 laboratory was disbanded because a number of VARs placed orders for it long after AT&T wanted to kill it off. It was handicapped in the market place by the fact that the 3B4000 Marketing Team was NOT ALLOWED to advertise it! This was in the era of Corporate Image advertising, and no individual product advertising was allowed. I remember leafing through a Telecom/Computing magazine one day and seeing full-page ads for our competitors' boxes, then I found a two page AT&T add featuring -- McDonald's French Fries (advertising the ISDN trial we were doing with McDonald's at that time). (-RS)

The Rainbow machines

Red, Orange, Yellow, Green, Blue, and possibly Purple, this was an exercise in futility which attempted to specify new machines each targeted for a wide variety of applications using emerging industry standards. Only the Yellow machine made it past concept stage, and development was scrapped after the requirements stage.

The Olivetti machines

Our first attempt to break into the Intel-based PC marketplace, with OEM machines from Olivetti. Though mostly run out of NJ, IW did systems engineering and test work on these. They were not 100% compatible with boxes from IBM, and did not sell well, at least compared to Compaq. They included the PC 6300, the 6300 PLUS, the 6312, and the 6320 WGS.

StarBASE

As the development of UNIX SVR4 was being planned by Summit, NJ, a historic partnership with SUN Microsystems was announced. SUN and AT&T were to develop SVR4 and after 6 months all the other UNIX licensees could have it. The UNIX world went bonkers and rebelled at this perceived plot to unfairly restrict the release of the software. The Open Software Foundation (OSF) was founded by the gang of seven (HP, DEC, etc.), and UNIX International (UI) was founded in retaliation by AT&T and SUN. Simultaneously, SUN announced a new RISC chip, the SPARC, and AT&T was to develop a large platform using this and UNIX SVR4 (SUN would continue in the workstation/small system side). The StarBASE was to be a multiprocessor of SPARC chips running SVR4 (kind of a forerunner of StarServer E, see below). The StarBURST was to be a unique crossbar-style networking facility using optical fibers which would allow different StarBASE "modules" to be spread around a building. Yet the entire StarBASE system was to appear as a single virtual machine, a la Apache. As the alliance with SUN was battered by market forces, the partnership eroded, and StarBASE was canceled before requirements were finished. AT&T later sold the Summit NJ facility and UNIX to Novell.

Cascade

The OEM PCs from Intel's Portland, Oregon operation, the Cascades were better known as the 6386 machines, with code names C3 and C4. Though it had a StarServer name, the StarServer S (codename C4E) was also a Cascade machine. Again, mostly run out of NJ, IW people did testing and application. The Skokie Teletype organization later moved to IW's second floor, bringing with them the C2 design they had originated (not OEM from Intel). By the way, there never was a C1 (figure that out).

Safari

AT&T's successful laptop computer, not just a PC, it was the first PC&C - Personal Computer and Communicator, with built-in modems and communications-oriented software. IW people did concept engineering and system engineering. The subsequent effort was run out of NJ and they were made in Japan.

E425M

The first attempt at a symmetric multiprocessor, it was to feature all Intel parts, including the Multibus-II. Only one was ever prototyped, and the machine was scrapped because of performance problems with the messaging architecture of the Multibus-II. Other companies working on Multibus-II eventually scrapped their models too.

StarServer E "Enterprise"

Potentially the most important system ever produced by IW, its life was cut short by the NCR acquisition. The world's second symmetric multiprocessor (only behind Sequent), and the world's first UNIX SVR4 SMP, it was fully 2 years ahead of the rest of its market (excluding Sequent) in its market niche. The system bus was designed to extend for 10 years through all foreseeable technology (like 100+ MHz Pentiums), and could still outperform the NCR 3000 Voyager bus six years after it was designed! It featured split-transaction memory operations, a bus arbitration mechanism which allowed 320 MB/s sustained transfers, dual interleaved memory structure, distributed interrupts, and a host of other hardware innovations to make it the most scalable SMP ever commercially produced by any vendor. It featured 33 MHz and 66 MHz Intel 486 versions, EISA bus for both 32-bit performance and compatibility with the IBM AT bus, and held the best price/performance slot in TPC Benchmark B for over a year. The regular model "Enterprise" could hold up to 4 CPUs and up to 2 GB RAM, and work on a bigger system codenamed "Bigfoot" would have offered up to 10 CPUs and up to 22 GB RAM, but it was never finished. Tricord Systems, headed by Jim Edwards who was once head of AT&T Information Systems, rejoiced the day NCR management canceled the SSE, because it opened up the market segment for their systems. The SSE was designed in to many Network Systems projects, and its cancellation by NCR management precipitated the shakeout which led to the ouster of the NCR old guard after the merger.

StarServer C

StarServer C was a desktop symmetric multiprocessor that followed the StarServer E and was in development when the desire to acquire NCR was announced. It provided either one or two 33MHz (or 25MHz) 486 processors, implemented a scaled down version of the E425M system bus, utilized the same custom silicon developed for the StarServer E memory controller, provided several EISA slots for compatibility, and ran the same OS as StarServer E. NCR was particularly befuddled by this system because they did not have a similar system on their roadmap and they realized from our market data that they should have, yet, they had to kill it because they did not think of it. Immediately after it was killed, a similar system appeared on their roadmap slated for availability 1 1/2 years later. Several StarServer C's were built and many found a home under peoples desks while IW was in the process of converting everyone over to PC's. (-BG)

System 7000

An OEM attempt to enter large scale database applications, this SMP system came from Pyramid Technologies, featuring up to 12 CPUs. Release 1 had a Pyramid proprietary CPU, and we pushed Pyramid into designing a MIPS-based CPU for Release 2 which they obtusely called "CPL" - customer provided logic. They have continued today with MIPS technology even as part of Siemens. IW people did specification and acceptance testing, and the system's biggest success was being installed into the billing operation for AT&T Universal Card, replacing IBM and Amdahl mainframes.

StarServer FT

An OEM project from Tandem Computers, this system was also totally fault tolerant (compare to the 3B20D), though it used a voting scheme in its CPU complex, but with dual active I/O paths. It featured a MIPS R2000 and later R3000 processor, and some of the software for the hardened UNIX kernel was developed by Bell Labs at Murray Hill, licensed to Tandem, which then delivered the finished UNIX kernel to IW. A unique demo was pulling out selected CPU, memory, and I/O boards while the machine still ran, and pulling out one of its three fans caused the other two to spin up to higher speed to take up the thermal load, again while keeping running. In a TPC Benchmark A audit, the auditor started a database transaction on the screen, the power plug was pulled, the system shutdown, the power plug put back in, the system came up, and the auditor hit the carriage return and committed the transaction! These were sold exclusively to internal AT&T applications, the biggest of which were BILLDATS and Autoplex. (-RJH)

The StarServer FT's were supplied eventually with R4000 and R4400 CPUs, the latter being 200MHz processors. They are presently used for the ANS (previously, AINET) line of Intelligent Network Elements (in addition to a product line based on Intel Pentium processors). (-RS)

INTREPID

A proposal to re-do "Bigfoot" with MIPS CPUs, it was never accepted by NCR management. It was part of the System 9000 proposal.

System 9000

A proposal to unify all MIPS-based IW machines into a market-cohesive single family image. The Phoenix would be the desktop, the Intrepid the main server, the SS-FT the specialty fault-tolerant system, and the 7000 the big machine. NCR management decreed nothing could compete with the NCR 3000 and promptly canceled all further IW system development.

LifeKeeper and DAP/TAP

At the NCR merger, part of the IW people were put into the NCR MidRange Computer Product Division. Reduced to being a software house to support Columbia SC, IW people ported LifeKeeper to bring High Availability Switchover technology to the NCR 3000 products. This unique software owns shared resources like disk drives and network connections, and automatically switches them between machines when a fault occurs. Unlike its competitor from HP, Service Guard, all machines in a LifeKeeper cluster are active, with only the applications being switched between them. Disk Array Plus and Tape Array Plus provided RAID-like technology in a software form. (-RJH)

LifeKeeper was originally developed for StarServers, and was working long before we merged with NCR. When ported to the 3000 products, it found all the bugs in the NCR port of SVR4. (-JCB)

Distributed Lock Manager and Oracle Parallel Server

This was a software project to make the NCR 3000 products the pre-eminent platform for this particular database product. The IW people performed Distributed Lock Manager (DLM) development and TPC Benchmark performance testing, and the 4-node cluster of 8-CPU NCR 3550s with 24 NCR 3350 clients was the biggest lab system assembled since Apache days. We succeeded at making the NCR 3000 the biggest OPS platform, but as this accounts for only 6% of Oracle's business, they continued to discount NCR as a partner.

Vistium

At the NCR merger, part of IW was put in the Workstation Products Division to help Clemson SC sell PCs. The major project was Vistium, a Windows-based application which allowed sharing of realtime audio, video, and application data between two PCs over an ISDN phone line. It was a leading edge product which has more capability than any other product even years later, but its dependence on ISDN combined with the closing of the NCR PC effort has scuttled the product.

Collaborative Work Board (CWB)

CWB was a separate effort in WPD at IW to provide a lower cost collaborative endpoint connected to a POTS line. It was a single ISA add-in card that featured V.32terbo (19.2K) modem with simultaneous voice & data, compatible sound blaster stereo audio, and child card provisions for a future low cost video codec. On the software side the product integrated the Vistium application sharing software, a telephone dialer and phonebook, caller-ID with distinctive ring, speakerphone, a full featured fax application, a telephone answering machine, and an audio applete gui (wav player/recorder, CD player and MIDI player). The CWB integrated all of the these features very tightly and offered features such as remembering audio level settings depending of the state of the telephone (e.g. when the local phone was taken off-hook, all audio levels would change to the user desired settings and would return when the phone was put on-hook). The CWB was about 2 years ahead of its time and was entering system test when marketing decided to cancel it under fears of having to support a non-standard SVD in the midst of an industry standard that was expected to be ratified within 6 months of product release. SVD did not become a standard until 18 months later. (-BG)

3B VME adaptor

The 3B product manager believed that if a VME bus was made available on 3B computers, we could instantaneously use tons of off-the-shelf VME peripherals. The project started around 1986-87 (I am not sure) and ended around 1988-89. The architecture was to develop an bus adaptor between the VME bus and the the Xbus (the extended bus which was developed for 3B15 in Apache). For the 3B2, a 3BIO bus to Xbus adaptor was developed. 3B20 later decided not to join the crowd. The first application was hyperchannel, and that was probably the only application. (-SJC)


Valid HTML 3.2!
Contents copyright 1996, 1997, 1999 by Michael S. Kenniston.

Last modified on 1999-05-28 by msk@xnet.com (Michael Kenniston).
Back to IW People on the Web page.
Up to Michael Kenniston's Home Page.