Systems of Systems and Emergent System Context

Paper #85 Systems of Systems and Emergent System Context

Nirav B. Shah Dr. Donna H. Rhodes Prof. Daniel E. Hastings Massachusetts Institute of Technology 77 Massachusetts Ave., 41-205 Cambridge, MA 02139 [email protected] [email protected] [email protected]

Abstract Introduction

One of the key issues facing architects of Since the mid 90’s there has been a grow- systems of systems (SOS) is the emergent na- ing interest in how systems come together to ture of the context of the SOS. In the spirit of form systems of systems (SOS). These tempo- Herbert Simon, context here means all the rary coalitions of independently operated and stakeholders, objects, processes and other ex- managed systems can meet unforeseen needs ternal effects that create the environment for in a timely and cost effective fashion. For ex- goal attainment of a system. As SOS are sys- ample, during the first gulf war a SOS was tems, they too have contexts. Understanding formed between Patriot missile batteries and the relationship between the contexts of the Defense Support Program (DSP) satellites. systems that constitute the SOS and SOS’s The satellites, meant for strategic missions, own context is a key step in building the archi- provided the Patriots’ operators with “over the tecture of a SOS. This paper will address SOS horizon” early warning of medium-range mis- context from several perspectives. First, con- sile launches prior to the inbound missiles be- text is defined and examples given to explain ing visible to the Patriot radar (Cunningham its relevance to system architecture. This is 1991). Combining disparate systems was not followed by a discussion of how the context of always so successful. During the lead up to an SOS is not simply the union of the contexts the same conflict, electronic coordination of its constituents, but rather emerges as a re- could not be fully established with naval sult of interactions that define the SOS. A key forces resulting in air tasking order needing to issue that arises from this view of SOS context be flown from Riyadh out to aircraft carriers is that the temporary nature of many SOS (Zinn 2004). There is much attention focused forces designers of constituent systems to be today on deliberately architecting systems so particularly adept at simultaneously operating that they support broader objectives of a SOS. in/planning for current, future and legacy SOS The first step is to identify those character- contexts in which they may be participating. istics that distinguish the SOS engineering The paper concludes with a discussion of dif- challenge from traditional systems engineer- ferent strategies for how constituent designers ing. Several authors have attempted such might handle this situation with recommenda- classification (e.g. Eisner 1991, Maier 1998, tions and examples drawn from historical Keating 2003, Gideon 2005). Though there commercial and defense SOS. are differences in how each author defines the SOS problem, three key characteristics stand out in their definitions:

1. SOS are systems the DSP satellites still needed to perform other 2. SOS are composed of other systems missions in addition to helping the Patriots. that are value producing in their own This paper focuses on collaborative SOS. right The reader should take SOS to mean collabo- 3. SOS constituents have some sense of rative SOS unless otherwise stated. One key independence after being assembled consequence of this independence is the com- into the SOS plex nature of the environment or context in which a SOS operates. SOS being systems, The first of these points indicates that SOS exist within an environment. Their constitu- are systems and as such possesses the charac- ents are also systems and so also have a local teristics of systems traditionally cited such as environment. Given managerial independ- stakeholders who build and derive value from ence, understanding the relationship between their use, an environment in which they oper- the SOS environment and the local environ- ate and, in the case of open systems, inter- ments of the constituents is important to creat- faces. The second point emphasizes that the ing successful SOS. This paper seeks to ex- constituents systems of a SOS are systems. plore that relationship and discuss architec- They too possess the system characteristics tural issues that arise from it. outlined above. In joining the SOS the constituents do not lose their identity as systems Defining Context and remain recognizable. The first two points alone, however, do not differentiate SOS as a While the term environment or context is class of systems requiring special attention – often used in the systems literature, there is given these points, SOS are simply systems disagreement on a precise definition. For ex- that happen to be composed of other systems. ample, some consider stakeholders inside the There are many systems that are composed of system boundary, while other consider them other systems. For example, a factory may be influencers on the system and therefore part of composed of several manufacturing cells, a the context. Some guidance in this regard can mainframe of several processors, etc. The be found in some of the early references. third point leads to the taxonomic distinction Ackoff (1971) defines the environment of a between SOS and the more general problem of system as: systems engineering. “The environment of a system is a set of Maier (1998) identifies two types of inde- elements and their relevant properties, which pendence experienced by constituent systems: elements are not part of the system but a operational independence (if removed from change in any of which can produce a change the SOS, a constituent could operate once in the state of the system. Thus a system’s again as an independent system) and manage- environment consists of all variables [that] can rial independence (even when incorporated affect its state. External elements which effect into the SOS, a constituent continues to de- irrelevant properties of a system are not part liver value independently). SOS that exhibit of its environment” these two independence criteria are termed Ackoff further notes that the environment ‘collaborative SOS’ and, as Maier argues, re- of a system can be subjective. Since its defi- quire a different architectural approach than nition is directly related to the system problem traditional systems. For example, in the Pa- being analyzed the same entity may be part of trior-DSP case cited earlier, full command and the system in one case and part of the envi- control authority over the DSP satellites could ronment in another. not be handed over to the Patriot operators, as Like Ackoff, Simon (1996) defines the ‘outer environment’ of a system as the ‘mold’

into which the successful system fits, i.e. the Context (architectural): The mold to environment, while not part of the systems, which the architecture must conform. It is a through its form and influence specifies the result of the objectives and constraints that the conditions or boundaries within which the sys- designer is trying to satisfy. tem achieves its objectives. Included are the An example of system context: A com- objectives themselves as they must be satis- munication satellite’s context consists of both fied for the system to exist – e.g. a plane must physical laws and requirements imposed by be able to fly for it to be plane. By tying the stakeholders. On the physical side, there is definition of a system’s environment to the the space environment and the realities of sig- objectives that its promulgators are trying to nal transmission physics, while stakeholders meet, Simon removes some of the subjectivity impose requirements on communication capa- of Ackoff’s definition. A key point of conten- bility and regulatory constraints on spectrum tion in the literature is if stakeholders are part usage. If an understanding of the entire prod- of the system or the context. Since this issue uct lifecycle is desired, then the acquisition often comes down to the semantics of how environment (e.g. schedule, budget, etc.) exactly an author defines the term ‘system’, comes into play. All these issues collectively there is no universally agreed upon choice. form the context that the spacecraft (program) For this paper, stakeholders are considered must fit to be successful. part of the system only if they directly partici- Since context defines the conditions for pate in the value creating action of the system. system success, understanding it is a key task If they only specify objectives or are affected for the system architect. However, as demon- by the output of the system, they are part of strated in the example above, context is often the context. Some authorities (e.g. INCOSE complex and multi-faceted. 2004) define the term environment in a nar- When one considers SOS, this challenge rower sense, i.e. the natural environment, becomes all the more difficult. regulatory environment or software environment, etc. The sense of environment being Context of SOS developed here encompasses all those environments in so far as they relevant to goal at- As specified in the definition, SOS are tainment for the system under study. Taken systems and, as such, have a context. This individually, however, they only provide one context specifies, to use Simon’s definition, piece of the environment as Simon defines it. the external conditions and influences that To prevent confusion this paper will use the must be dealt with for goal attainment. From term context for environment in its broader the analytic perspective, determining the con- sense. Based upon Simon and Ackoff, two text of a SOS is a similar problem to determin- definitions of context are provided below. ing context of any complex system. Architec- When considering a system already in exis- ture frameworks can be helpful in managing tence, the analytic definition applies, while the data collection process and much literature design of new or modification of existing sys- exists on their use (see, for example, Bernus tems calls for the architectural definition. 2003). A more difficult problem arises when Note that the two definitions specify the same one considers context from a design perspec- entities as being part of the context. tive. Context (analytic): The external entities As an analogy to the SOS case, to see how and conditions that need to be taken into ac- context develops when two systems interacts, count in order to understand system behavior. consider Figure 1.

map overlayed with rental opportunities (Rademacher 2005). This new entity, Hous- ingMaps (www.housingmaps.com), is a SOS as defined above. See Figure 2 for a sample screenshot of listings for downtown Boston. The constituent systems remain quite independent in both an operational and a managerial sense. They are so independent in fact that neither Craigslist nor Google participated Figure 1: Two Systems operating sepa- directly in Rademacher’s endeavor. At the rately, but sharing context time of development there wasn’t even an API available for Google Maps, Rademacher had Two systems, A and B, are generically to rely on underlying web technologies (Goo- represented as circles. They are each situated gle 2005, later, spurred on by effort such as in their respective context indicated by the Rademacher’s, an API was released). Since blue and green rectangles respectively. They HousingMaps is a system, it has a context. are not interacting directly, but do share some This new context is represented in Figure 3. context. For example, system A could be the rental property listing service on the website Craigslist1 and system B, Google’s map serv- ice2. In this example, the shared context includes a common technical foundation, i.e. the Internet and WWW, and some common stakeholders, e.g. users of both services. These users soon found out that these services can be used together – find a property to ones liking on Craigslist and then map the address in Google Maps to get a sense of neigh- Figure 3: Context of a SOS borhood. The two services being independent however, did not facilitate such collaborative Unlike what one might naively be led to use. One user, Paul Rademacher, noticed that believe, the context of the SOS is neither the there were tools available to make a Google union nor the intersection of the contexts of its

1 Craigslist is a website (boston.craigslist.org) that provides listings of user submitted classi- fied advertising similar to local newspapers. 2 Google Maps (maps.google.com) is a web- based inteFraiguctirvee 2m: aHpou sesrviincgMe frapoms sGhooglowineg. Craigslist rental listings on Google Maps

constituents. Rather, it includes entities from ponents that are related only to it are irrelevant both constituents’ contexts and also introduces when considering the constituents in isolation. new element/constraints that do not apply One of those contextual components is the when considering the constituents in isolation. need or requirement for the SOS behavior. Applying this diagram to the HousingMaps This need is not part of the constituents’ con- example, one first notices that the website acts text. Therefore, the SOS’s context contains at as an interface between the two constituent least this element that does not exist in the systems3. In doing so a new context is cre- constituents’ contexts and will probably con- ated. The SOS context inherit the Internet ar- tain other elements that arise from the SOS chitecture and related constraints from its con- need. stituents, and also has new constraints levied The situation becomes even more complex by the technologies used by Rademacher in when one allows for constituents to participate developing the website. For example, the in multiple SOS. Both the constituents in the scripting engine used to develop Housing- HousingMaps example participate in other Maps is not relevant when considering the SOS. In fact meeting the needs of Housing- constituents in isolation, but become very im- Maps is a minor objective of those constitu- portant with looking at the SOS. This would ents. Contextual constraints may arise from lie in the teal area in Figure 3. Conversely, these other SOS engagement and impact the some portion of each of the constituent’s con- SOS under study. The quick rise of sites such text is not relevant to the SOS. For example, as HousingMaps placed great strain upon Craigslist must abide by government fair Google’s infrastructure and they were forced housing regulations – such regulations do not to put limits on information provided to these impact HousingMaps. Given the example of external sites in order to maintain quality of HousingMaps, it seems that the context of the service (Google 2006a). This constraint on SOS can be complex and is not easily deter- geo-location services arose from neither Hous- mined by looking at the contexts of the con- ingMaps’s nor Google’s actions alone, but stituents alone. was a consequence of the overall popularity of Unfortunately this situation is not limited the services Google Maps provided. Figure 4 to the example given above. Rather it will be extends the previous context diagrams to in- typical for SOS because they are systems. As clude this effect: systems they exhibit emergence in the sense of Simon (1996), i.e. they exhibit behavior that is distinct from the union of the behavior of the constituents. Simply bringing together several systems does not form a SOS. One must have them interact to produce value that could not be obtained without said interaction to form a system and thus a SOS. Since this new value producing behavior did not exist prior to forming the SOS it follows that any context com- Figure 4: Possible cases for participation of a constituent in multiple SOS 3 One can also interpret this as a three entity Figure 4 shows the many possible SOS in SOS: Craigslist + Google Maps + the Hous- which system ‘A’ could be participating. The ingMaps backend and UI. In this view, Goo- horizontal axis indicates that at the same time, gle and Craigslist are interfacing with Hous- system A may be participating in multiple ingMaps and not with each other. SOS (the AB and AC SOS). In addition, there

is potential for future participation is the as yet the characteristics Maier proscribes for a SOS, unrealized AD SOS. The architects of each of ensuring that the individual systems could still these SOS may need to aware of one or more operate independently. of system A’s other SOS since they may im- Much work went into planning the integra- pact their own architecture. For example, if tion of the many dozen systems involved in the same system A (as opposed to copy of A) the AWE. Architectures were developed, pro- is simultaneously participating AB and AC, tocols were documented and constituent level the pressure on systems A’s resources from test plans were implemented. Despite all the AC interaction on AB ability to make use these efforts, the initial attempt at integrating of system A’s resources is an important part of the SOS went quite poorly. The many sys- the context of AB. Contextual issues may also tems simply did not interoperate correctly spring up when one considers SOS evolution given demands being place on them. Systems over time. In Figure 4, the AD SOS may be that worked in isolated test failed to scale hobbled by the legacy interface A need to when tried in the full exercise. maintain with C. Despite the best intentions of the develop- In summary, determining the context of a ers of the individual systems, integration sim- SOS is a non-trivial exercise for the SOS ar- ply did not work. They had demonstrated that chitect and will require consideration of issues it is very difficult to, a priori, understand the beyond those identified in the context of the shared context formed when an SOS is assem- constituents. The context of the SOS emerges bled without actually forming the SOS. Their as a result of the interaction between the con- difficulties were compounded by the fact that stituents. Other SOS in which the constituents several of the systems being integrated were participate may also have an impact. Given new or were operating in new ways. This lim- this challenge, what then is the SOS architect ited their knowledge of constituent contexts to do? Fully answering that question will re- and thereby made understanding and staying quire more research into the mechanisms of within the limits of the SOS all the more chal- how the SOS context emerges, however his- lenging. To resolve this impasse, the Task torical examples can provide some guidance Force XXI team gave authority to their ‘Cen- to the architect. tral Technical Support facility’ to manage the integration of the SOS. The leader of the Creating context: SOS Integration CTSF was given the authority to arbitrate between the needs of the various constituent sys- Given the widespread interest in SOS tem stakeholders with the intent of producing there have been surprisingly few in-depth case a cohesive whole. The result was transition studies of SOS. However, a few good case from a haphazard integration effort in which studies do exist and can provide the architect the SOS context was constantly shifting as with guidance on how to manage the complex systems tried to integrate, to an orderly inte- context of the SOS. One particular study of gration process driven by SOS needs. This note is Krygiel’s (1999) examination of the authority also reduced the length of decision Army’s Task Force XXI Advanced War making cycles and allowed rapid spiral devel- Fighting Experiment (AWE). The AWE was opment of the SOS capability. an attempt by the Army to determine how a The Task Force XXI story represents one networked brigade would function. The intent approach to managing the complexity of SOS was to bring together the latest in information context – strong central authority to arbitrate technology with restructured tactics and then and prioritize system evolution and spiral de- gauge their effectiveness through a series of velopment to allow for rapid learning and elu- war games. The integrated system did have

cidation of the SOS context. The downside of understand how SOS context comes to be, the this approach, however, is that the SOS as- literature seems to indicate that centralized sembled is the one conceived by the integrat- coordination may be a good way to keep inte- ing authority that may or may not be the best gration complexity at a manageable level. solution. Such control, however, will limit innovation The integration experience summarized and prevent SOS like HousingMaps from de- above is quite different from that which pro- veloping. Finding the right balance between duced HousingMaps. In the HousingMaps control of the integration process and open- case, the constituents were not even aware of ness to allow for creativity of the constituents there participation until integration was well is a major challenge for SOS architects. on its way. More importantly, the existence of robust interfaces or ‘hooks’ in the constitu- References ents’ architecture meant that the constituents did initially not need to know about the SOS Ackoff, R.L., “Towards a System of Systems integration. However, once the SOS ham- Concepts”, Management Science, v.17:11, pered the constituents’ ability to meet other July 1971 objectives, corrective action became neces- Bernus, P., Nemes, L., Schmidt, G., Handbook sary. Had Rademacher taken an approach on Enterprise Architecture, Springer, 2003 more like Task Force XXI, such difficulty Cunningham, J.H., “The Role of Satellites in could have been avoided. Resource usage the Gulf War”, The Journal of Practical could have negotiated upfront instead of dealt Applications in Space, v.2:3, Spring 1991 with once the crisis was upon the constituents. Eisner, H, Marciniak, J, McMillan, R, “Com- In the end finding the right heuristics for man- puter-Aided System of Systems (S2) En- aging the emergence of SOS context will re- gineering”, IEEE SMC, 1991 quire further study. Google has moved to INCOSE, Glossary in System Engineering model more like Task Force XXI in the next Handbook, Version 2a, June 2004 version of their API terms of use requesting Maier, M.W., “Architecting principles for sys- that if “your site gets more than 500,000 page tems-of-systems”, Systems Engineering, views per day, we ask that you talk to us be- v.1:4, 1998 fore you launch so that we can prepare in ad- Keating, C., Rogers, R., Unal, R., Dryer, D., vance to handle your traffic.” (Google 2006b) Sousa-Poza, A., Safford, R., Peterson, W., Such a system ensures that smaller sites can Rabadi, G., “System of Systems Engineer- grow without the need of Google’s attention, ing”, Engineering Management Journal, yet allows Google to work proactively with v.15:3, Sep. 2003 larger sites. Krygiel, A., Behind the Wizard’s Curtain, CCRP, 1999, see chapters 3-4 for a de- Summary scription description of the case and chap- ter 5 for lessons learned from the integra- This paper has defined and developed the tion experience. concept of system context when applied to Gideon, J.M., Dagli, C.H., Miller, A., “Tax- SOS. The context of an SOS is complex en- onomy of Systems-of-Systems”, CSER, tity the results from the interaction between 2005 the SOS constituents and may not be readily Google, “Maps/Craigslist mashup”, Google understood by looking at just those constitu- Code Blog, 2005, ents. Environments in which constituents are http://google-code-featured.blogspot.com/2005/04/ forming multiple SOS further complicate the mapscraigslist-mashup.html situation. While more research is needed to

Google, “Google Maps API Terms of Use”, Past-President and Fellow of the International 2006a Council on Systems Engineering (INCOSE). http://www.google.com/apis/maps/terms.html Daniel Hastings is a Professor of Aero- Google, “Google Maps API Version 2”, nautics and Astronautics and Engineering Sys- 2006b tems at MIT. Dr. Hastings has taught courses http://googlemapsapi.blogspot.com/2006/04/google and seminars in plasma physics, rocket pro- -maps-api-version-2.html pulsion, advanced space power and propulsion Rademacher, P., HousingMaps, 2005 http://www.housingmaps.com/about.html systems, aerospace policy, technology and Simon, H.A., The Sciences of the Artificial, 3rd policy, and space systems engineering. He Edition, MIT Press, 1996 served as chief scientist to the U.S. Air Force Zinn, A., The Use of Integrated Architectures from 1997 to 1999 and as director of MIT’s to Support Agent Based Simulation, Mas- Engineering Systems Division from 2004 to ter’s thesis, Air Force Institute of Tech- 2005. He is a member of the National Science nology, Wright-Patterson Air Force Base, Board, the International Academy of Astro- OH, 2004. nautics, the Applied Physics Lab Science and Technology Advisory Panel, and the Air Force Biography Scientific Advisory Board.

Nirav Shah is a graduate student at MIT pursuing a Ph.D in Aeronautics and Astronau- tics. A member of the V-STARS research cluster and supported by MIT's Lean Aero- space Initiative, his research focuses on architecture of systems of systems. In particular, he is studying the impact of coupling in the physical, functional, and organizational spaces on system evolution. His work experiences include positions at Los Alamos National Laboratory and with Booz Allen Hamilton. Nirav received an S.B. (2001) degree and an S.M. (2004), both in Aeronautics and Astro- nautics, from MIT. Donna Rhodes holds a Ph.D. in Systems Science from the T.J. Watson School of Engi- neering at SUNY Binghamton. Her research interests are focused on systems engineering, systems management, and enterprise architecting. Dr. Rhodes has 20 years of experience in the aerospace, defense systems, systems integration, and commercial product industries. Prior to joining MIT, she held senior level management positions at IBM Federal Sys- tems, Lockheed Martin, and Lucent Tech- nologies in the areas of systems engineering and enterprise transformation. Dr. Rhodes is a