CASE STUDY
`
`EXPERT SYSTEMS FOR CONFIGURATION
`AT DIGITAL: XCON AND BEYOND
`
`Members of Digital Equipment Corporation’s team of expert system experts
`reflect and recount a decade's worth of lessons learned in designing and
`building a core of configuration systems
`
`VIRGINIA E. BARKER and DENNIS E. O’CONNOR
`
`The XCONconfiguration system at Digital Equipment
`Corporation was thefirst expert system in daily pro-
`duction use in industry [2, 10, 11]. It is the cornerstone
`of Digital’s knowledge network vision: a numberof ex-
`pert systems embeddedin both the company’s order
`process cycle and its new productintroduction cycle
`[4]. Today, Digital is continuing to extend the knowl-
`edge network, as well as using expert systems technol-
`ogy in manyadditional aspects of the company’s busi-
`ness. In fact, XCONis only one of several expert
`systems dealing with hardware and software config-
`uration which are currently in use or under develop-
`mentat Digital. The configuration systems “family”
`includes four expert systems in production use. Several
`additional configuration expert systems are in the
`research, advanced development, or prototype stage of
`development.
`The development of these expert systems has
`spanned nearly 10 years; Digital’s Configuration Sys-
`tems Development Group (CSDG) has thereby gained
`considerable understanding ofall phasesof thelife-
`cycle of production quality expert systems: design, de-
`velopment, production, and ongoing support.In this ar-
`ticle we highlight some of the key lessons we have
`learned:
`
`¢ Building a successful expert system involves much
`more than simply putting rules into a knowledge
`base. Rather, to successfully develop and provide
`ongoing support for expert systems andto integrate
`them into the fabric of one’s business, as has been
`doneat Digital, one must attend to the needsof the
`business and to humanresource and organizational
`issues as well as to technical issues.
`e Although many software engineering practices
`carry over to the engineering of expert systems, we
`
`© 1989 ACM 0001-0782/89/0300-0298 $1.50
`
`have nonetheless neededto be creative in develop-
`ing practices specifically tuned to this relatively
`new,andstill evolving, technology.
`
`While the accompanyingarticle by Bachant andSolo-
`way focuses specifically on the technical issues under-
`lying XCONandtheother configuration systems, we
`take a moreholistic, integrative approach in this article
`and attempt to put the aforementioned concernsinto a
`coherent perspective.
`
`CONFIGURATION SYSTEMS AT DIGITAL TODAY
`The Digital Configuration Systems Timeline (p. 302)
`shows,for each of the configuration systems, key stage-
`of-life milestones, including initial production use
`whereappropriate.
`XCONis usedto validate the technical correctness
`(configurability) of customer orders and to guide the
`actual assembly of these orders.It provides the follow-
`ing functionality:
`
`¢ Configures CPUS, memory, boxes, backplanes, cab-
`inets, and power supplies, disks, tapes, HSC/CI,
`printers, etc.
`e Diagrams complete system configuration (Figure 1
`overlays selected pages of a sample XCON diagram)
`e Checks marketing restrictions, system building
`block menus, and prerequisites
`e Assigns addresses/vectors and determines box
`powerstatus
`e Partitions multiple-cpu orders and cluster systems
`e Determines andlists cabling information
`e Lists components ordered with configuration-
`related comments
`e Generates warning messagesonissues affecting
`technicalvalidity.
`
`XSEL [7, 9] is used interactively to assist in the selec-
`tion of saleable parts which make up a customerorder.
`
`298
`
`Communications of the ACM
`
`March 1989 Volume 32 Number 3
`
`CONFIGIT 1044
`
`CONFIGIT 1044
`
`1
`
`

`

`Case Study
`
`It provides the following functionality:
`
`e Allows interactive selection by generic component
`name, partial or full model number
`e Performs completeness checking, adding and sug-
`gesting required parts
`e Checks software compatibility, prerequisites,li-
`cense and media completeness
`¢ Checks standard and system building block menus,
`marketing and engineering restrictions
`* Provides computer room environmental data and
`requirements
`¢ Links to Automated Quotation System.
`
`XFL is used to diagram a computer room floor layout
`for the configuration(s) under consideration. It provides
`the following functionality:
`
`e Provides “minimum footprint” floor layout of com-
`ponents
`e Allows custom rooms(user-specified dimensions
`and placement)
`¢ Can include several configurations or a cluster in
`onesite layout.
`
`XCLUSTERis usedto assist in configuring clusters. It
`provides the following functionality:
`
`¢ Clusters multiple-node configurations for validation
`¢ Specifies device quantity for dual porting
`e Upgrades appropriate clusterable tapes.
`
`In addition to these four systems currently in produc-
`tion use, two other configuration systems are under
`development:
`XNETis an expert system which will be used to
`design local area networks, to select appropriate compo-
`nents for such networks, andto validate the technical
`correctness of the resultant network configurations.
`SIZERis a research effort addressing the need for
`tools to assist in the sizing of computing resourcesre-
`quired for any of a wide variety of uses in various types
`of organizations.
`To support the ongoing developmentof the configura-
`tion systems in production use and to enable us to more
`effectively build new configuration systems, we have
`developed a software engineering methodology,called
`RIME[1, 12, 13], expressly for expert systems. Because
`RIMEprovides substantial insight into the engineering
`issues that require particular attention in building ex-
`pert systems, it is discussed at some length in “The
`Engineering of XCON.”
`
`SCOPE AND USAGE
`The configuration systems provide full product cover-
`age for Digital’s current product set. This productset
`today consists of 42 different families of central proces-
`sor types and their supporting peripherals and software.
`In order to be useful business tools, released versions of
`these systems must include configuration knowledge
`of Digital’s newest products by the time of product
`announcement. In practice, this means that CSDG
`provides major releases of these systems once each
`
`quarter, with at least one interim upgradeto insure
`adherenceto the time-of-announcement requirement.
`Hardware andsoftware configuration is at the core of
`Digital’s business. The configuration systems are used
`worldwide, throughoutthe corporation, by a broad set
`of users across the company’s major functions: sales,
`manufacturing,field service, and engineering. The
`users of these systems perform functions which span
`Digital’s complete order flow and manufacturing cycle,
`and,thus, are involved with manydifferent business
`processes. This is a large and varied constituency to
`support—eachhasdifferent needs and takesa different
`perspective on the configuration information provided:
`
`e Sales uses the configuration systemsas an integral
`part of the automated process to generate quota-
`tions for customers, and to insure that every order
`is technically valid.
`¢ Manufacturing uses the information to verify build-
`ability of all incoming orders, to understand phys-
`ical partitioning of an order into various sub-
`assemblies to determine which plants should build
`which segmentsof an order, to guide the assembly
`of all orders, to determine the optimalset of diag-
`nostics to run on eachorder.
`¢ Field service has the perspective of consolidation
`and assembly of the order in the customer’s unique
`environmentandpossibly with existing equipment
`already installed.
`e Manufacturing and engineering benefit from the
`configuration systems’ focus on system integration,
`as analysis of product knowledgefor inclusion in
`the configuration systems identifies potential prob-
`lems in system-level design and manufacturability.
`
`This user profile has expanded dramatically over the
`years (see timeline). For example, the initial purpose of
`XCONwasto assist manufacturing plant personnel in
`validating the technical correctness of system orders
`aboutto be built. Since then, this technical validation
`function has changed in responseto business needs and
`is now performedin thefield as well. In addition, the
`technicians in the plants who build the computers now
`use XCONdiagramsto see how to put the systems
`together, and they include those diagramsin the ship-
`mentto the customerfor use byfield service installers
`at the customersite. XSEL wasoriginally designed for
`use by sales representatives, and is now used by our
`OEMcustomersas well. Implementation of XNET will
`add specialized field support personnelto the userlist
`in the near future. There are additional “indirect” users
`of these systems through automatedlinkages to other
`software systems(both traditional and expert systems)
`which depend onthe configuration information
`supplied.
`The configuration systems user base not only repre-
`sents over 10 distinct business functions, but is also
`spread across the world, and has varied geography-
`specific needs. Overall, CSDG supports over 50 produc-
`tion installation sites, and the numberis growing. In
`sum,then, the configuration systemsare firmly em-
`
`March 1989 Volume 32 Number 3
`
`Communications of the ACM
`
`299
`
`2
`
`

`

`Case Study
`
`
`
`1'||l|
`
` IDEC-NUMBER 6200TSTO1 CUSTOMER-NAME CALYPSO PROD-LINE 0
`
`
`DEC-NUMBER 6200TSTOL
`CUSTOMER-NAME
`YPSO PROD-LINE 0
`
`
`RUN DATE/TIME:
`7?-DEC-1988 08:12
`
`H9646-SA
`#2
`
`
` XCON RELEASE AS OF VS.4-0
`
`
`FULL SYSTEM CONFIGURATION { WITH ADDITIONS AS REQUIRED }
`
`
`OMPONENTS ORDERED
`
`NAME,
`DESCRIPTION
`COMMENT
`
`
`
` 62ACA-DP
`6210 CLUST SYS 208/60 PPL
`
`KAG2A-AA
`PROCESSOR
`
`
`32 MEGABYTES OF MEMORY
`
`2
`5
`DMB32-M
`MULTIFUNC VAXBI, SASYN, 1SYN,L
`
`3
`5
`DHB32-M
`16-CHAR ASYNC MUX,VAXBI,W/MC
`2 OF THESE WERE NOT
`CONFIGURED
`4
`4
`DRB32-E
`DRB32-M + EXTERN ADP NO CABL
`
`5
`4
` DRB32-W
`DRB32-M + DR1i-W ADP NO CABL
`1 OF THESE WERE NOT
`CONFIGURED
`
`6
`2
`TUS1E-BA
`TU81-PLUS, BI-BUS,
`60HZ
`1 OF THESE WERE NOT
`CONFIGURED
`
`7
`1
`8c00B-aAC
`STAR COUPLER, 8 NODE WITH CAB
`
`a
`1
`TA79-BF
`DUAL ACC MASTER 240V 60H2
`9
`2
`SA600-HA
`4.8GB SA BUILDING BLOCK, 60H
`
`
`OMPONENTS ADDEDNAME
`COMMENT
`
`
`DESCRIPTION
`
`NEEDED TO CONNECT
`BC26V-12 SHIELDED SI CABLE, 12FT.
`BSCSX-BA AND
`
`
`
`FRONT-A~O00sO
`
`
`a
`2
`Bc26v-12
`SHIELDED SI CABLE, 12FT.
`NEEDED TO CONNECT
`HSC5X~BA ANDOgangsaborenccrene rccoettc
`
`
`
`
`FRONT-A-1
`
`(FRONT)
`(BACK)
`
`
`
`
`
`
`12
`2
`BCZ6V-12
`SHIELDED $I CABLE, 12FT.
`NEEDED TO CONNECT
`
`
`
`
`HSC5SX-BA AND
`$4600 DISTRIBUTION PANEL ASSEMBLY
`FRONT~A-2
`
`
`
`
`
`SHIELDED SI CABLE, 12FT.
`BC26V-12
`1
`13
`74-36165-03
`SIGNAL~DISTRIBUTION
`1
`NEEDED TO CONNECT
`HSCSX-BA AND
`
`
`T
`t
`MI
`T
`“T
`FRONT-A-3
`
`
`C1780 CABLE-10 METERS
`BNCTA-10
`2
`14
`HJ
`| a6
`j A2 HSCSX O
`|| AZ HSCSX 1
`| AT
`t
`NEEDED TO CONNECT
`!
`!
`a
`1
`|
`HSC70-AA* AND
`
`
`| BE
`| B2
`tt B3
`|
`!
`SCO08-AB
`
`
`
`4s CK-DHB32-AJ=CAB KT RS232/WMC, VAX85XX-8803° NEEDED FOR DHB32-M [eennnreesernnn [erewennnneceee P[nnecrerence-sen l
`
`
`
`
`
`
`
`
`
`
`
`16 CK-DMB32-LJ-CABKIT SFT CBLE VAK#800/85005 NEEDED FOR DMB32-M $ AO HSC5x O | A4 {| AS | Al HSCSX 1 J
`
`
`
`
`
`
`
`
`
`4 CK-DRB32-Lg=SFT CABLE FOR DRB32 NEEDED FOR DRB32-E* I t I 4 '
`
`
`
`
`
`
`
`
`
` DEC-NUMBER 62007STO1 CUSTOMER-NAME CALYPSO PROD-LINE 0
`
`
`1
`DWMBA-BA
`6200 VAXBI_ EXP CAB 208/60
`REQUIRED BY TBK50
`| BO
`\ B4
`\ BS
`! Bl
`\
` CABINET LAYOUT
`
`
`
`
`fonerennn-ane]
`[TO79-ar
`170-24900-01 (H9657-EC
`yCAB #1
`
`| TUS1E-CA 0
`
`REQUESTOR SLOTS
`
`0123456769
` DEC-NUMBER 6§200TSTO1 CUSTOMER-NAME CALYPSO PROD~LINE 0
`
`
`70-24900-01 #0
`
`
`
`3
`3h 31 34 3)
`
`
`eten- Hnae|
`
`
`(UNIT # 0
` |HSC70-AA 0
`
`
`.CH DRB32-W REQUIRES A CK-DRB32-xX CABINET Kit TO BE ATTACHED TO A USER-DEVICE
`NSIDE THE CABINET
`
`OMPONENTS NOT CONFIGURED
`
`
`2
`DHB32-mM
`16-CHAR ASYNC MUX, VAXBI,W/MC
`MARKETING RESTRICTION
`
`1 DRB32-W
`DRB32-M + DR11-W ADP NO CABLE
`CANNOT BE CONFIGURED WITHOUT ADDITIONAL BOX SPACE
`
`
`
`TUB1-PLUS, BI-BUS, 60HZ 1 TU61E-BA
`NO CONTROLLER WAS CONFIGURED FOR THIS DEVICE
`ON ANOTHER CONFIGURATION,
`TEMS NOT CONFIGURED AND ARE TO BE USED AS SPARES,
`
`DR FOR TESTING
`
`
`OPTION UNIT
`IDWMBA/A
`4
` 12-23701-01 BI-PLUG INTERCONNECT-HARDWARE
`
`
`
`
`
`DWMBA/A
`5
`12-23701-03 BI-PLUG INTERCONNECT-HARDWARE
`IKAG2A-AA
`0
`
`12-23701-08 BI-PLUG INTERCONNECT-HARDWARE
`
`12-23701-09 BI-PLUG INTERCONNECT~HARDWARE
`
`42-23701-10 BI-PLUG INTERCONNECT-HARDWARE
`12-23701-11 BI-PLUG INTERCONNECT-HARDWARE
`
`12-23701-12 BI-PLUG INTERCONNECT-HARDWARE
`
`12-23701-13 BI-PLUG INTERCONNECT-HARDWARE,
`12-23701-14 BI-PLUG INTERCONNECT-HARDWARE
`12-23701-15 BI-PLUG INTERCONNECT~HARDWARE
`
`CABLE ASSY, FLEX. B/P EXTERIO
`17-0103B-01
`17-01170-04
`CABLE BUNDLE
`
`17-01170-05
`CABLE BUNDLE
`17-01462 CABLE
`17-01462-01
`DRB32-W
`BULKCHAD ASSY,
`70-23923-01
`
`
`NODE-ID1
`BC22D-25
`5
`A
`
`
`
`
`
`
`eBURYHPUWUWELUUUES
`
`2
`17-01897-01
`FROM DWMBA/A_5 CONNECTORS K1J2-D2/D1/8
`CONNECTORS K7J1-D1/D2/E1 /E2 IN CAB ¢
`
`
`LENGTH:
`15 FEET
`LENGTH-REQUIRED:
`15 FEET
`2
`17-01897~03
`FROM DWMBA/A 1 CONNECTORS KiJ13~-D2/D1/82/
`
`CONNECTORS K3J1-D1/D2/E 1/B2 IN CAB # 0
`
`LENGTH:
`2 FEET 1
`INCH
`LENGTH-REQUIRED:
`2 FEET 1 INCR
`
`FROM DRB32-E* 3 CONNECTORS K2J3-C2/D2/E:
`3
`17+02375-02
`D5-D6 IN CABINET 0 -- U SING CK-DRB32-:
`LENGTH:
`$ FEET
`LENGTH-REQUIRED:
`2 FEET 6 INCHES
`
` UNUSED CAPACITY
`
` HSCS5X-CA 0 CAN SUPPORT
`
`HSC5X-BA 0 CAN SUPPORT
`
`HSC5X-BA 1 CAN SUPPORT
`1 BC26v-12
`FROM HSCSX-BA 0 TO FRONT-A-0 ON SI
`Hoot
`band past tot tent
`TS78-BB 0 CAN SUPPORT
`LENGTH:
`12 FEET
`
`HSC70-AA* 0 CAN SUPBPOR’
`
`
`2113 1
`14
`7
`15
`1
`Wt
`12113 1
`14
`|
`a5 1
`16d
`LENGTH-REQUIRED:
`12 FERT
`tant tent dont
`
`
`FROM RASO-AA* 8 ON HSCSX-BA 1 TO BACK-A-
`2
`17~01482-01
`
`
`SIGNAL-DISTRIBUTION PANEL 2
`
`LENGTH:
`6 FEET
`LENGTH-REQUIRED:
`6 FEET
`FROM HSC70-AA* 0 TO PATH-A NODE 0 IN SCOOB~g
`2 BNGI-10
`
`LENGTH:
`32 FEET 10 INCHES
`LENGTH-REQUIRED:
`32 FEET 10 INCHES
`1 BNCI-20
`FROM CIBCA-BB* 0 TO PATH-A NODE 1 IN SCG
`
`LENGTH:
`65 FEET 7 INCHES
`
`LENGTH-REQUIRED:
`32 FEET 10 INCHES
`
`
`2
`70~17382-15
`FROM TS78-BB 0 TO TU79-AF* UNIT $ 0
`LENGTH:
`15 FEET
`LENGTEH-REQUIRED:
`15 FEET
`
`1 BC17x-20
`FROM KLESI-BB* 0 TO TUBIE-CA* 0
`LENGTH:
`20 FEET
`
`LENGTH-REQUIRED:
`6 FEET
`
`1 BC2év-25
`FROM HSC70-AA* 0 PORT A3 TO TS78~BB 0
`LENGTH:
`25 FEET
`12 FEET
`LENGTH-REQUIRED:
`
`
`FIGURE 1. Selected Pages from a Sample XCON Diagram
`
`3
`
`

`

`Case Study
`
`beddedin Digital’s most critical business processes. The
`existence of these expert systems hassignificantly im-
`proved these processes and continuesto provide insight
`into opportunities for the future.
`
`BENEFITS
`Only a modest amountof attention has been given to
`documenting the benefits of applying expert systems
`technology [3, 8]. The configuration systems are a suc-
`cess and there is a major dependency on them within
`the corporation, worldwide. They benefit Digital in a
`numberof ways, contributing to customersatisfaction,
`lowercosts, and higher productivity. These systems
`are recognized asa critical factor in Digital’s ability to
`maintain its highly successful a la carte, build-to-order
`marketing strategy (customized configurationstofit
`each customer’s specific needs); this is one of the com-
`pany’s key competitive advantages. Some of the bene-
`fits are difficult to quantify, but overall the net return
`to Digital is estimated to be in excess of $40 million per
`year.
`Theuse of the configuration systems insures that
`complete, consistently configured systems are shipped
`to the customer. Incomplete orders do not get through
`the process. In addition, XCON generates configurations
`which optimize system performance, so customers con-
`sistently get the best view of our products. Before the
`configuration systems, we would often ship the same
`parts configured differently. (There are multiple ways
`to configure the sameset of parts to create a working
`system.) This was a major source of customer com-
`plaints and confusion, especially for OEM’s whoorder
`large numbersof the same system andin turn recon-
`figure them all in the same wayas part of their market-
`specific value added process.
`The process of new productintroduction is enhanced
`greatly by the focus on configuration information pro-
`vided by the configuration systems. The existence of a
`single source of configuration information by the time
`of product announcementfor use on initial customer
`orders simplifies field and manufacturing training
`needs and avoids confusion about new products which
`can delay time-to-marketsignificantly. This is of criti-
`cal importance given the volumeof new products Digi-
`tal continues to introduce each year. This single source
`of configuration information also increases manufactur-
`ing’s flexibility by enabling product manufacture (and
`the knowledgeof howit is done) to be moved from one
`plant to another without costly training or disruptive
`re-assignment of people. The use of XCON hasfacili-
`tated this re-alignment of manufacturing capacity sev-
`eral times.
`The use of the configuration systems hassignificantly
`increased the technical accuracy of orders entering
`manufacturing. Straightening out problem ordersis a
`costly and time-consuming activity which disrupts the
`normal processes and increasesorder cycle time(i.e.,
`delays order shipment). Overall, the additional disci-
`pline of both field and manufacturing use of these sys-
`
`temshas shortened cycle times, contributed to more
`flexible, smoother-running processes, and lowered the
`numberof people who would otherwise be neededfor a
`given volumeoforders.
`XCONis seen as a critical componentof Digital’s
`current process of shipping segments of an order from
`various componentplants, to consolidation points, and
`from there to the customer site. The use of XCON
`throughout the manufacturing process assures that
`when the components of the order cometogether for
`the first time at the customersite the system will work.
`Before the use of XCON,there were special manufac-
`turing plants where every customer order was com-
`pletely assembled, tested, disassembled, and repacked
`prior to shipping in orderto insure the system would
`work wheninstalled at the customersite. The elimina-
`tion of this step in the manufacturing process(called
`Final Assembly and Test) has had a major positive im-
`pact on cycle times, inventory levels, and manufactur-
`ing costs, especially in light of Digital’s four-fold in-
`crease in systems product volumesince 1980.
`Thus, the configuration systemsare key in Digital’s
`ability to handle the product complexity that technol-
`ogy advances are enabling and that our customers are
`demanding.
`
`THE CONFIGURATION SYSTEMS EXPERIENCE
`AtDigital, we believe that our success in the applica-
`tion of expert systems technologyis attributable to a
`conscious recognition of and careful balance among
`three important perspectives: strategic/business, tech-
`nical, and human resource/organizational.
`
`STRATEGIC/BUSINESS ISSUES
`Digital has reaped and continues to reap enormousben-
`efits from the configuration systems. It was, of course,
`an experimentaleffort at first but it was nurtured and
`allowed to progress and turnedoutto be a resounding
`success. Now Digital considers its use of expert systems
`technology a strategic investment.
`Whatare the characteristics of an appropriate busi-
`ness problem for this type of strategic investmentin a
`new technology? The problem mustbe real and sys-
`temic to the enterprise to justify the right kind of nur-
`turing and supportive business environment. Thesolu-
`tion will probably impact multiple organizations or
`functions and, hopefully, will bring about significant
`and positive changes in the way the enterprise oper-
`ates. The configuration problem at Digital certainly
`qualifies. Digital’s strategy of selling customized solu-
`tions is oneofits critical competitive advantages, and
`gives rise to an indefinite numberof valid configura-
`tions of its product set. In the 1970s, problems with
`handling this were starting to surface, and it was
`understood that automated support would be critically
`important as sales volumes increased and products
`became morevaried and complex.
`Oneof the main challenges we have encountered
`in developing systems which supportstrategic, cross-
`
`March 1989 Volume 32 Number 3
`
`Communications of the ACM
`
`301
`
`4
`
`

`

`
`
`TWWLIDIG
` azigdnoip
`
`LoerrSIEMPIEH/AIEMYOSGaJOOMIONadeSLOISNO)heeen(jC)a/HUISuoneunbyuo5
`
`
`
`
`
`
`=OLE———.001'SZ008PLGeen00901ade09'S00S(19Ge()()5————————00!yuauodwio9
`00821SE009'FLG002101001002Lg06'S0020580080S082joJaquinn
`————theeEHEErtSe——01ES©|#aBe1aA00
`
`———6Staoise———C—O22Z-———!!juawdojessg
`
`mowersUONONPOIdGulAjoAy4Padojanapguepadojarsp—Bunjong——""0J001y
`
`
`
`
`
`SSUUUEEEAEEEEERnenEEnEEEeaEeeeEooj‘saibayesSuneysew‘seiunsojouemau‘uoneinBas904—@=sasnqMan
`
`
`
`
`
`6861886t28619861S861vs86l£861c86L186LO86L6261
`
`smox/13x/73asxsmoxWXTaSXx
`BULANUOreHUANGsSILIS[]aIAIDUODJOJOOId
`
`
`
`uoljonpod3WiyBursnuoyonpoidadAyoyod
`
`
`swayshsuononpoid389}playydaqu09
`
`
`wia)sksABojopouyewwa}shs
`
`
`
`AguapuadapiajulsndoSMdd40wa}shsuoMUagySey
`
`
`
`
`
`/msosegisApouenD—$—$—=$—=—_$—$_—$——fuaueno+=oy——_—_—=—_—_—_—_——_einOiouquandaryyuawidojanaq
`
`
`
`
`
`sepes6dnwyajutsasegjaiseseajalsaseajesssaooid
`
`Y3ZIS
`
`sadAjo\oid
`
`
`
`
`
`ysOLajqeesnydaou05
`
`
`
`
`
`ANMSWILSWALSASNOLLVYNDISNOD
`
`errsP/O}Gece11)/NIN@10jo01g
`
`L3NX
`
`wa)sks
`
`182}pial
`
`paw]
`
`
`
`NOOX34dJOA9VLS
`
`
`
`
`
`ewes§}90U09JOJOO
`
`
`
`s1a}snjo
`
`3ZISGNV3dO00S
`
`Awe}NdO
`
`
`
`abesaaooeyep
`
`ulsued#)
`
`{aseqeyep
`
`(suuaysAs
`
`We(2303)
`
`sain
`
`AIN3SW35VNVAW
`
`5
`
`

`

`Case Study
`
`functional business needs, is sorting out how the tech-
`nologists connect with the business. As with anysoft-
`ware developmenteffort, it is necessary to form a solid
`partnership with the business being supported. Usually,
`that means with the end users and their immediate,
`operational management. In the case ofstrategic, cross-
`functional systems, we have discovered that the devel-
`opmenteffort must also be tightly connected at a more
`strategic level of management. This is simply a reflec-
`tion of the fact that the problem being addressed has
`strategic impact and the solution may include and/or
`provide opportunities for far-reaching changes in the
`business [6] of which the tactically focused end user
`groups may be unaware.
`CSDGis tapped into business planning at several lev-
`els to insure adequate awareness and connectivity and
`support. This has been and continuesto beofcritical
`importance to the ongoing success of the configuration
`effort at Digital. Much of the strategic connectivity over
`the years has been informalfor the configuration sys-
`temseffort: based on personal relationships with key
`individuals. Recently, we have initiated someefforts to
`formalize someof these processes. For example, we
`haveset up a Configuration Systems Steering Commit-
`tee madeupofstrategically focused managersrepre-
`senting our various business constituencies. This forum
`provides guidance on planningandpriority-setting for
`the configuration systems in the contextof a cross-
`functional vision looking out several years. In addition,
`a more formalcorporate-level focus on system integra-
`tion issues is evolving, which should provide formal
`long-term direction.
`Oneof the most important factors in our success has
`been the existence of an enabling business /develop-
`ment environmentfor the introduction of new technol-
`ogies like expert systems. During the early 1980s, Digi-
`tal’s manufacturing operations were open to embracing
`newideas and technologies that would help them
`change the way work was done,affording higher pro-
`ductivity and shorter cycle times in response to cus-
`tomer requirements. Further, it was believed that this
`expert system technology could help shape the way
`Digital did business, allow the companyfunctions to
`manage complexity, preserve the knowledgeofits ex-
`perts, and provide an approach to exception manage-
`ment more consistent with its growth demands. An
`environment has been provided which has nurtured
`the emerging technology and recognized that the peo-
`ple involved are managing change, evolution, andsig-
`nificant learning. It is understood that there will be
`new and changing operating normsand successcriteria
`over time. Thus, at Digital there has been real manage-
`ment commitmentto helping the technology succeed in
`helping the business progress.
`
`TECHNICAL ISSUES
`Technologies develop iteratively through a numberof
`phases:
`
`e Investigation and experimentation,
`e Stabilization with limitations,
`
`e Identification of opportunities for optimization, and
`e Maturity.
`
`Noaspects of expert systems technology havereally
`reached maturity yet. This need not be an impediment
`to its successful use, but it must be understood and
`proper expectations mustbeset. In a new and emerging
`technology, there are many opportunities for technical
`creativity as new discoveries and capabilities surface.
`Thereis also the likelihood that a serious application
`developmenteffort will push at the limits of what the
`technology hasto offer at a particular point in time, and
`in so doing, can contribute greatly to the further devel-
`opmentof that technology.
`
`Technology Challenges
`The main challenges stemming from the emerging
`nature of this technology are:
`
`1. Dealing creatively with the current limitations of the
`specialized languages andtools, and
`2. Formulating an appropriate software engineering
`methodology(this is the thrust of the RIME develop-
`menteffort)
`
`to produceefficient, maintainable, extensible expert
`systems.
`Today’s expert systems languages (andshells) offer
`varied but somewhatlimited functionality. Most do not
`provide true extensibility to a production environment.
`There are very few knowledge engineering productivity
`tools available as yet on the market. The OPS5 lan-
`guage wasoriginally chosen for the developmentof R1,
`the forebear of XCON. As XCONandtheother configu-
`ration systems have evolved using VAX OPS5,® its
`strong pattern-matching capability and its speed have
`maintained it as our language of choice overall others
`on the market today. CSDG hasdirectly influenced the
`development of VAX OPS5, based on our experience,
`and continuesto do so.
`Bachant and Solowayintroduce a software engineer-
`ing methodology appropriate for the developmentof ex-
`pert systems. They make twoprimarypoints:(1) A task
`such as XCON’s, where thousandsof highly context-
`dependent decisions must be madein orderto correctly
`configure a system, can take advantage of a language,
`like OPS5, that presupposes “situation recognition”
`rather than “algorithmic” control. (2) But, as the num-
`ber of alternatives that an expert system must decide
`among becomeslarge,there is strong motivation to aug-
`mentsituation recognition with other kinds of control
`mechanisms. RIME, a software engineering methodol-
`ogy developed within CSDG, identifies three mecha-
`nisms that an OPS5 program canuse to determine what
`to do and whento doit. This methodology, which was
`first used to reimplement XCON,has proven beneficial
`in the managementof complexity and facilitation of
`change.
`
`®VAX OPS5is a registered trademark of Digital Equipment Corporation.
`
`March 1989 Volume 32 Number3
`
`Communications of the ACM
`
`303
`
`6
`
`

`

`Case Study
`
`Testing, by traditional standards, is a major problem
`for expert systems. Can you test an expert system the
`way you evaluate a human expert? Since the under-
`standing of the business problem is often enhanced
`over time by the use of the expert system, how can you
`define a set of tests that are sure to be relevant from
`one release to the next? Since there are, typically, an
`indeterminate numberof paths through an expert sys-
`tem, it is not possible to exhaustively test all scenarios.
`Whatis correctness? Consider the relationship between
`the optimal solution and an accurate solution. An error
`can be difficult to distinguish from a wish-list item. The
`expert system may comeupwith the right answer, but
`for the wrong reasons. Whataboutthe cases where the
`experts disagree? There are other related testing issues
`as well, all of which cry out for new viewsof design-
`for-testability and new testing concepts, methods, and
`tools. This is assuredly an area which needs moreat-
`tention, and we hopeto devote more energyto this in
`the future.
`
`Application Challenges
`There are three application characteristics which are
`major contributors to the challenges faced in expert
`systems development:
`
`1. Volatile subject domain,
`2. Expanding functional scope, and
`3. Large system size and complexity.
`
`These are primary characteristics of the configuration
`systems, and have presented us with our toughest chal-
`lenges over the years: (1) computer configuration is a
`dynamic subject domain;(2) once the systems were
`viewed as successful business support tools, users have
`wanted more and more from them; and(3) the configu-
`ration systems have becomevery large and are dealing
`with increasingly complex information relationships.
`Success has comeonly from ourability to respond to
`these challenges without disrupting or degrading the
`accuracy of the systemsor ourpredictability in provid-
`ing regular production releases.
`
`Volatility
`Each year, about 40 percent of the rules in the configu-
`ration systems knowledge base change(this includes
`rule additions and deletions, as well as rule modifica-
`tions). This rate of change in the engineering, manufac-
`turing, and marketing rules is due to a numberoffac-
`tors. Digital offers several hundred major new products
`and many more “minor” ones each year. Configuration
`information about them all must be included in the
`configuration systems knowledge base. In addition,
`sometimes knowledge about how to configure existing
`products must be changed to incorporate their configur-
`ability with new ones. Also, existing products are often
`re-packaged,as dictated by engineering, marketing, or
`governmentregulation. And,finally, CSDG periodically
`re-writes focused sets of rules as more is learned about
`
`the configuration of a particular product set and how to
`represent the knowledge moreeffectively.
`There is another dimensionof volatility in the con-
`figuration domain: the scope of the configuration task
`itself has changedsignificantly. The complexity and
`flexibility of Digital’s product strategies has increased
`steadily since XCON wasfirst designed. A single com-
`puter system can no longerbe configured in isolation,
`but must be consideredin relationto its role in a clus-
`ter or network. Clusters and networksoffer multiple
`waysof configuring computers and connecting devices.
`Cluster and network connections are coming to be
`thoughtof as the main bus, with systems and devices
`providing functionality for one large “solution system.”
`Moreover,distinctions between systems and devices
`are becomingless clear. The configuration-related in-
`terdependencies between hardware and software are
`much more numerous and complex today. In summary,
`the basic concepts and componentsof the configuration
`domainare being redefined continuously.
`
`Expansion
`The configuration systems we havenotedservein-
`creasingly varied user groups from different business
`functions and geographies. New types of users with
`new perspectives generate new requirements which
`frequently enlarge the functional scope of the systems.
`For example, when the plant technicians began to use
`the XCONdiagram asthe official document from
`whichto construct the computer, a whole new set of
`requirements was generated. CSDG madeanarchitec-
`tural change to the system in order to support these
`needs(altering some of the knowledge representation,
`transitioning functionality between the OPS5 rules and
`non-OPS5 codeto create the diagrams, and developing
`a specialized database).
`Astheir businesses change, existing users have dis-
`covered new anddifferent ways to use the configura-
`tion systems. For example, as order volumes havein-
`creased andthe technical] validation function has
`changed, those users wantto interact with the tool in
`different ways and need different kinds of information
`from the systems. They have asked for more stream-
`lined ways to access the XCONresults, as their pro-
`cesses have changedto precludethe needfor steps pre-
`viously performed andreflected in the XCON menu
`access scheme.
`When XCONstarted feeding information to other
`systems, there have been instances wheretherules
`have had to be changed to accommodate a newper-
`spective on the configuration information. For example,
`there is another expert system used in manufacturing
`whichhelps judge howtosplit the construction of the
`orderacross several plants which specialize in certain
`sub-assemblies [4]. This system needs information from
`XCONonthe physical partitioning of the order and
`whatparts are used in what sub-assembly.XCON rules
`hadto be changedto be able to explicitly provide this
`information at an intermediate point in XCON’s overall
`pro