Distributed Systems Architecture
written by Jean Bellec from Claude Rolland ."Bull et les Communications" (covering also pre- and post- DSA )
In the late 1970s CII-Honeywell-Bull and Honeywell Information Systems undertook an important effort to design a network architecture competing with IBM SNA, more flexible and taking advantage of the, then existing, open standards.
Honeywell had drafted a preliminary specification of a distributed architecture called HDSA (Honeywell Distributed Architecture) that had for goal to give some meaning to an expected revamping of Series 60 product line that was re-announced in 1978 with a DPS suffix. It was thought as an alternative to IBM's SNA offered from what still "the other computer company".
The CII-HB participation to that design was inspired by the previous CII investment in NNA (New Network Architecture) that significantly closer to market than HDSA. Both, the CII-HB and the Honeywell designers were quite active respectively in ECMA and ANSI standardization committees designing the ISO/OSI standards (Open System Communications).
That effort was given the name of DSA Distributed System Architecture; it would encompass the strict functions of networking and was designed to be extended in distributed data bases and applications.
Goal of DSA.
For Honeywell and CII-HB, DSA was an effort
complementary to the Series 60 systems and GCOS(es) to be able to offer to their customers
a total offering of an information processing solution.
This total offering only excluded pure telecommunications equipment that were, at that time, supplied on an exclusive basis by monopolistic common carriers (AT&T, European PTTs).
DSA was a head-on alternative proposal against
IBM Systems Network Architecture that was considered as a tentative of establishing a
total monopoly by IBM.
DSA was designed on top of some PTTs networking standards (such as HDLC, X-25), but was not offered to competitors or other players of the industry for several reasons:
Honeywell (and CII-HB) were still in the 1970s considering themselves as "The other computer company" and did not want Univac, Burroughs and DEC take benefit of their R&D efforts.
DSA was advertised very early vis-à-vis the state of developments and standards were not ready for publications and if they had been published Honeywell and Bull had no lead-time advantage over the competition.
Official standards (OSI) were expected to emerge momentarily and the DSA strategy was to negotiate the compatibility of those standards with ISO/OSI and to replace DSA with OSI through a minor update release of the DSA products.
The introduction of DSA was expected also to contribute to the rationalization of R&D efforts and to make credible the unification of GCOS disparate product lines. By allowing the coexistence and centralizing the administration of heterogeneous computers, DSA should allow a more graceful migration (either by recentralizing or by decentralizing applications).
DSA would also give an opportunity to accelerate the internal migration of customers inside the GCOS product lines, pressing for getting out of BCD GCOS III into ASCII GCOS8 and acquiring new hardware.
DSA was lead by a distributed team of architects working in Louveciennes (near Paris) and Phoenix AZ. The Paris team was lead by Claude Boulle and in Phoenix, among the main architects were John Couleur and Charlie Bachmann.
Implementation of the products were also distributed in separate departments: Network front-end software in Louveciennes, GCOS8 software in Phoenix, GCOS6 software in Billerica MA, GCOS7 software in Paris. Impact of DSA on GCOS4 was minimal and nil on MULTICS.
The critical DSA products were not covered by licenses granted by Honeywell to NEC, and were considered as an umbrella to the NEC possible hostile entrance in the world-wide market.
DSA strongly adheres to ISO layered-OSI architecture.
Physical links supported by DSA were serial interface to modems (for WANs) and Ethernet thick interface for LANs.
Link layer .
DSA primary links specify HDLC protocol, used as point-to-point links.
DSA adopted X-25 packet switching protocol such as adopted by almost European PTTs in the 1970s.
DSA chose to use the connection option on top of X-25 and having an end-to-end guarantee of the reliability of exchanges.
The connection-less mode would have been often more convenient to large transaction applications, and the rigidity of the model was circumvented by ad-hoc mechanisms in TP subsystems.
The presentation layer was very far from external standardization in late 1970s and was sketchy in DSA.
DSA standards addressed:
File Transfer and remote Batch
The development of DSA products was marked by a lot of small controversies between implementers (responsible for performances and tempted to collapse layers, or to distribute functions according the availability of processor power), product planners (wishing to minimize DSA impact on customer existing applications and hardware) and architects (wishing to keep networking functions inside DNS to allow their eventual evolution towards OSI).
The key product of DSA is an unified front-end and network processor (UNCP). It was used as the data communications front-end processor for GCOS8 and GCOS7 processors. It was to replace existing front-ends on GCOS8 and upgrade integrated communication processor on GCOS7.The front-end processor kept the name of Datanet (inherited from GE times).
The initial DSA Datanet (DN71xx) was a Level 6 processor designed in Billerica and manufactured in Boston and Joué-les-Tours (France). Several models with increased performances were later introduced. The last Level-6 based model was Datanet 7500 (1991) multiprocessor version of MRX/A.
The software, called DNS (DataNet system) was developed by CII-Honeywell-Bull was a stand-alone real-time operating system, completely independent from GCOS6.
The front-end processor connects to GCOS8 processors trough the same DIA interface as its predecessors, allowing a limited support of DSA functions (terminal emulations) , coexisting with GRTS interfaces on Level66 existing processors.
The connection of DPS-6 to Level64 and DPS-7 is done through a new PSI channel adapter, developed by Billerica on Louveciennes specifications. The support of that adapter from GCOS7 side was designed and programmed by the GCOS team (FNPS Front-End Processor Support)
The Datanet was also available as a stand alone network processor.
In 1985, Bull had developed on a 68000
processor a smaller version (called internally MicroFEP of a front-end processor (for
GCOS7 DPS-7000) and it modified the DNS software factory to produce indifferently Level6
and 680x0 code. A progressive phase-out of Level-6 proprietary hardware was undertaken.
Eventually, this project converges into the MainWay new front-end product announced circa 1993 to support simultaneously DSA and TCP/IP networks.
The world-wide Datanet park peaked at 4500.
In addition to GCOS7 and GCOS8 "mainframes", a DSA network includes GCOS6 processors that runs GCOS6 software that was complemented by the support of DSA interfaces, protocols and products. That DSA software was designed and implemented in Billerica.
DSA concepts were also including dumb terminals that were guaranteed uniform support by network processors and mainframes applications.
Two standards of terminals were initially supported:
Character Level teletypes including "glass typewriters" and micro-computers (compatible PCs and Questar 400 CTOS computers) using terminal emulation.
VIP terminals initially developed in Honeywell
Oklahoma City. They use an Honeywell proprietary block level protocol. The most known
being the VIP7800, manufactured in Boston.
CII-HB developed its version of VIP terminals as Questar 200. A VIP7800 emulator on IBM compatible PC was also delivered (by Phoenix) and is still used today to support VIP7800 dependent applications.
Some effort was paid to design a standard device independent protocol to be used exclusively by applications. Although successive versions were implemented in some products, this effort cannot called a success. .
GCOS DSA Basic Software
GCOS7 implemented session control in a product called VCAM Virtual Communication Access Method, providing a centralized control of the communications functions for GCOS applications (file transfer, transactional, programs directly interfacing with terminals through COBOL verbs...)
GCOS DSA Middleware
Remote Job processing
GCOS DSA Applications
Coexistence with IBM SNA
.In the early 1980s, SNA had conquered an important part of Honeywell and Bull customers, the pre-1970s world of segmented worlds (IBM and non-IBM) having vanished. Bull attempted for some time to have IBM abandoning its grip on SNA by regulations at EU (then CEE ) level. IBM accepted to make available its standards at announcement time.
Bull started an ambitious program , called Janus,
that was to allow the connection of a DSA subnetwork to IBM mainframes. The Janus program
was to develop in Datanet a 370x (IBM front-end) hardware and software interface to IBM
VTAM software and to support LU0, LU2 SNA protocols in that Datanet. This program was
successfully implemented and several French customers implemented mixed networks.
A fall back of the strategy was to support LU2 terminals (3270 family) as part of DSA. The 3270 support opened DSA networks to many more terminals manufacturers than the Honeywell/Bull evolving VIP standards.
Independently from the physical networks unification, mixed customers require telecommunications exchanges between IBM and Honeywell/Bull mainframes. An initial solution was provided in the early 1980s by connecting GCOS mainframes to a Network Control Systems (later acquired by StorageTek) Hyperchannel. Hyperchannel support was not incorporated into DSA and remains a special product.
In IBM coexistence, file transfers were the most in demand, but some distributed transaction applications (with CICS and IMS/VS) were also demanded. So, GCOS7 TDS and GCOS8 TP8 implemented a Distributed Transaction protocol functionally equivalent to (and mappable on) IBM LU6.2.
In the early 1980s, the introduction of microcomputers had no negative impact on the DSA, it allowed to expect the generalization of intelligent terminals and to solve the problem of presentation control (for instance, through SGML). The Bull strategy was still to maintain the DSA philosophy, to keep it a closed system until the generalization of OSI.
Bull (in liaison with Honeywell) had been extremely active in the promotion of ISO standards and under some customers request (such as EDF wishing to switch from its proprietary RETINA system) started to implement ISO.
That strategy was disturbed by the universal acceptance of TCP/IP as a de-facto standard and by spreading of Internet. GCOS systems, as SNA systems, had to scramble to cope with the newly established standards and the ISO/OSI initial implementations were to be considered as specials. DSA implementations remained supported in the 1990s (and probably in the 2000s) but in coexistence with the TCP/IP world.
Revision : 11 juillet 2001.