When I teach architecture classes, I increasingly run into architects who never architect a “system”. They do all the things architects do: analyse stakeholder needs, identify ways to address these needs, (help) decide on the best way by making trade-offs, communicate these decisions and oversee the solution’s implementation. However, after all the work is done, the stakeholder needs are fulfilled, but there is nothing in the real world that they can point to and say: “that is the thing I architected”. At the end of the road, the implementation of their solution to the stakeholder needs consisted of a series of changes to existing systems, but did not require the creation of a new system. Their job is to architect transient solutions: changes to a domain or product they are responsible for.

One example was brought to me by the architect responsible for a transportation ticket vending machine. There was a stakeholder need to reduce the time and cost required to deploy new products and user interaction schemes, and the solution was to move most of the logic and data from the vending machine to a central place. Projects like this require significant (and careful) design effort, and substantial implementation time. A documented solution architecture is a must. However, at the end of the road, no new systems have been created (with the possible exception of minor components like hubs between previously unconnected systems).

Depending on the prevailing flavor of delivery governance, such transient solutions are defined in architecture epics, project documents or change requests. The architects designing them call themselves “domain architect”, “project architect”, “tech lead” or “master builder”.

What’s in a name

Is this really architecture? The ISO 42010 definitions of architecting and architecture both refer to a “system” of interest, e.g. “fundamental concepts or properties of a system in its environment embodied in its elements, relationships, and in the principles of its design and evolution”. All the other words seem to apply, so can we ignore the fact that there is no one “system” to point to after implementation of the architecture? Is this question really relevant, as long as the person designing these solutions knows how to apply architecting principles and practices?

If this is architecture, what type of architecture is it? Of all the different genres of architecture (software, infrastructure, system, enterprise architecture, etc.) the term solution architecture seems to best cover these transient solutions. This phenomenon actually helps us to create an appropriate definition for the term solution in solution architecture. Where previously we had to make do with something like “a solution is a way to solve a problem”, I would suggest the following:

Solution: a coherent set of changes delivered to address a defined set of stakeholder needs.

The changes referred to in this definition pertain to elements which can be newly created, modified or removed as part of the solution. The term coherent implies that the (changes to the) elements cooperate to achieve the fulfillment of the stakeholder needs (using Fred Brooks’ term, the solution has conceptual integrity). Coordination of a solution’s delivery depends on the governance model applied, this can be using agile or traditional delivery models, in a value stream, in a program or project, governed by a contract or otherwise. How and where the stakeholder needs are defined also depends on the governance model applied, this can be in an epic or set of (user) stories, but also in a program or project definition, contract or change request. From the perspective of this definition of solution, architecting a new system or service is just a special case where the “set of changes” consist of the creation of one new system or service.

Applying architecture practices

How does this way of looking at solution architecture affect the way we apply architecture practices? To explore this question, let’s take Risk- and Cost-Driven Architecture (RCDA) as our reference practices. As stated in the introduction, most core practices apply to both transient solutions and to system (or system of systems) solutions:

As you can see from this brief analysis, the main differences when applying architecture practices occur after the solution has been delivered. The most visible difference is what happens to the architecture documentation, for which it makes no sense to remain “as a unit” after implementation of a transient solution, except for historic reference. Extra care must be taken that the relevant information (such as rationale of architectural decisions) remains available afterwards, either in the documentation of the systems affected, or at a higher level (e.g. in domain or enterprise architecture documentation).

Conclusion

As less and less green field systems are created, an increasing number of architects find themselves mainly designing transient solutions – solutions that dissipate into the landscape and cannot afterwards be identified as a product or system. Good architecture practices are still applicable in this situation. This realization leads to the idea for a defininition of Solution Architecture as the “architecture of a coherent set of changes delivered to address a defined set of stakeholder needs”.

Architects in the digital world often compare themselves to architects in the construction world. The metaphor works on many levels: architects in both worlds are responsible for conceptual (and structural) integrity, are the content leaders with overview over design and construction, making key design decisions and drawing blueprints. There is at least one aspect, however, in which the construction world is very different from the digital world: IT-based solutions such as software, infrastructure and services are subject to change far more frequently than buildings.

The role of change in the digital world is recognized in the ISO 42010 definition of architecture, which explicitly mentions evolution as part of architecture. Change is also the central theme in modern software development methodologies like Agile and DevOps. The digital architecture disciplines seem to be lagging behind a bit in this development, perhaps hindered by the metaphor that gave them their name. It is time to give change and evolution their proper place in the digital architecture world, and make Time a first class concept for architects of software, infrastructure, services, enterprises etcetera.

Issues with time-agnostic architectures

When construction of a building is finished, its architecture documentation is final: it will not change for decades. The architecture documentation was probably considered final before construction began, since in the vast majority of cases only non-architectural changes occur during the construction of buildings. It is perhaps this idea of “final and unchanging architectural documentation” that prompts managers to ask architects “when will your architecture be finished”? In view of the prominence of change in the digital world, however, solution architects can rarely give a straightforward answer to this question.

Nowadays, practically all software-intensive systems are part of a complex application landscape, forming systems-of-systems with myriads of interdependencies between commercial and bespoke software systems, hardware platforms and organizational entities, all with their own evolution cycles. In such a landscape, a time-agnostic architecture is a very perishable good: its best-before date is at most weeks in the future. This  leads to the following observed issues:

• architecture documents that are perpetually “almost finished” (causing delays in projects dependent on them) or already obsolete when they’re issued
• development based on obsolete architectural assumptions
• difficulty planning ahead

Popular architectural styles like SOA and microservices attempt to reduce the pain caused by these ever-changing dependencies at the technical level, but can provide only limited relief when it comes to logical dependencies. In modern application landscapes, the only other way for architects to address the issues observed above seems to be to design the solution’s evolution into the architecture. We have to create architecture documentation that not only describes the as-is and to-be situations, but explicitly identifies and deals with architecturally significant events on the way from as-is to to-be.

Architecting Time: the Evolution viewpoint

According to the ISO 42010 standard, architecture documentation consists of views that represent the architecture from certain viewpoints. Viewpoints are designed specifically to demonstrate to stakeholders how the architecture addresses a particular set of their concerns. Philippe Kruchten’s seminal “4+1 View model” paper gives five good examples of viewpoints that do this for common stakeholder concerns in software development. What if we added an “evolution viewpoint” that is specifically designed to show how the architecture addresses the impact of changes in the solution’s environment? An Evolution view would identify future events that will have architectural impact on the solution. It would specify that impact in terms of business value, cost and risk, and analyze dependencies between the solution and the events. The view would also document how the solution (and/or its delivery/operations team) will anticipate and react to the events when they occur. Typical examples of events with architectural impact are:

Stakeholders interested in the Evolution viewpoint would be anyone who has concerns related to change and planning: specifically project/program/product managers, product owners and  architects/designers/developers working on other solutions in the same interdependent system of systems. It would help the managers and product owners plan ahead, and by acknowledging future events in the time dimension it would help fellow workers that depend on your architecture by telling them what aspects of the architecture will change, and when.

In short

Making the Time dimension part of your architecture documentation may look like more work, but anticipation should be a large part of your job as architect anyway. By writing it down (for example using an Evolution viewpoint), your architecture description will stay valid longer, and you will have a ready answer when stakeholders ask you how their change and planning concerns are addressed.

Sources:

Software Systems Architecture – Working With Stakeholders Using Viewpoints and Perspectives by Rozanski and Woods, specifically the Evolution perspective.

Enabling Agility Through Architecture by Brown, Nord and Ozkaya gives us the tools for “Informed Anticipation” in the architecture: dependency analysis, real option analysis and technical debt management.

DevOps is an approach to IT that radically alters the traditional relationship between development, maintenance and operation of software. The traditional separation between Dev and Opsbrings with it a substantial amount of waste in terms of elapsed time and budget. This waste is caused by differences in way of working, attitude and hand-over inefficiencies. The waste is exacerbated by accountability mechanisms that incentivize Dev and Ops departments to erect barriers, to protect themselves in case the other side fails to fulfill its responsibilities (“cover your behind”).

In DevOps, this waste is largely eliminated by making one team responsible for both DEVelopment and OPerationS of applications. Supported by extensive automation (mostly by open source tooling) of testing and deployment, DevOps allows organizations like Facebook, Netflix and Google to deploy new releases into the Cloud on a daily basis. This is quite an improvement over the usual “once or twice a year” releases seen in traditional software products, and it gives organizations tremendous agility in reacting to market developments.

DevOps appears to be most successful in end-user facing (front-end) applications that run in public or private clouds and have a web-based user interface. The tooling supporting the approach is fascinating, especially in the way it ensures quality control: see Netflix’s Dianne Marsh take on this tooling at this year’s Saturn conference. But how does DevOps impact the architecture of the software that is developed with it? So far, I have found three ways in which DevOps influences architecture:

• Architecting for deployability. Of all the quality attributes that a system may have, the one that is most directly related to DevOps is the ability to quickly deploy new releases into operation, with as much automation as possible. As I already mentioned in a previous blog, Stephany Bellomo and Rick Kazman of the SEI have made a study of this, and so has Len Bass of NICTA. An important consequence for architects appears to be the necessity to simplify and streamline the design. This includes standardization of infrastructure components, but also removing architectural elements which were originally intended to ensure other quality attributes, now overruled by the higher priority of deployability. One remarkable example of streamlining the SEI researchers found was the removal of an Enterprise Service Bus in one organization: originally implemented to enhance connectivity and modifiability, the ESB was found to be an obstacle to the speed of (automated) deployment and removed in the transition to DevOps (just an example – not a tip, ESBs generally have no quarrel with DevOps).
• Architecting for systemic resilience. As the speed and scale of deployment increases, chances are that from time to time stuff (services, app, APIs) that your software depends on may not be available for a few minutes or hours. Your software needs to be able to cope with this: the system as a whole needs to still be as available as possible. Your particular piece of the solution needs to stay alive, even if the whole world turns against it. Organizations like Netflix harden their system (and I guess their developers!) by actually making things worse: their simian army of chaos monkeys, latency monkeys and so on wreak havoc in test and production(!) environments. As Netflix says, “the best defense against major unexpected failures is to fail often”.
• Consolidated development architecture. In software development the target architecture for a solution has always put constraints on the set-up of the development environment, but with DevOps, it becomes almost impossible to see the development and operational environment as separate entities. From an architectural point of view, the development and target environments form one system with three main categories of users: developers, operators and end-users. One could even argue that that first two of those categories are the same in the DevOps philosophy. Many quality attributes affect all these user categories, and many architectural decisions have impact across the board as well. In DevOps, an architect (or “master builder”) does not design a target architecture after which an appropriate development toolset is selected: the whole ensemble forms one consolidated solution, with one solution architecture that addresses the combined architectural concerns for development and operation.

So as an architect, you may need to learn three new things: 1) to remove stuff rather than add stuff, 2) to design failure into your solution and 3) to design a consolidated development environment into your architecture. And you had better learn fast: Gartner calls this architectural approach “Web-scale IT”, and predicts that by 2017 50% of global enterprises will be using it.

What other aspects of architecture do you think is impacted by DevOps and Web-scale IT? Please leave your comments and let me know what I missed.

When I teach solution architecture classes, technical debt is always a very popular topic among practicing architects. Technical debt is a metaphor that transposes the concepts of loan and interest to IT based solutions. It respresents work that should be done in order to deliver a consistent, maintainable solution. As long as the work has not been done, the solution is in debt, which means that some stakeholders pay interest in the form of e.g. extra effort needed for simple changes, or higher support fees. Repaying the loan’s principal means doing the work needed to remove the debt: this could mean e.g refactoring software or upgrading hardware. As soon as that work has been done, the stakeholders stop paying the interest, just like when a loan has been repayed.

Architectural debt

The technical debt metaphor has been very popular in the software development world, where it refers mainly to low code quality or unnecesary complexity. Tools like SONAR now have functionality that analyses source code to measure such “implementation debt”. For architects, however, other types of technical debt may be more interesting. Aside from implementation debt, there is “architectural debt”: this is typically structural in nature, or represents a technology gap.

An example of structural architectural debt is when an architectural principle like “all applications should use the Enterprise Service Bus (ESB) to exchange data” is temporarily violated. An architect could decide to allow direct access from application A to application B’s database if A needs data of B that have not yet been exposed through the ESB, and doing it properly would mean missing an important deadline. The interest in this case is caused by reduction of control of the information flow through the application landscape, and potential errors being introduced by teams that are not aware of the shortcut. The principal is the refactoring that needs to be done later on: changing applications A and B to route the data through the ESB, and configuring the ESB. This type of debt cannot be measured in the software code of either application: it is structural in nature. The same is true of technology debt,  when a solution uses obsolete hard- or software products that cause potential failures and risks (interest) and needs to be upgraded (principal).

Business case

One of the problems many architects face is convincing their stakeholders to reduce technical debt, mainly because the debt is invisible to the end-user (see Philippe Kruchten’s categorization). Making the case for technical debt reduction in technical terms will usually not convince the business stakeholders. That requires translation into economic terms – in other words, a business case.

If one has unlimited resources (time, budget, staff), the business case for repaying technical debt is quite simple: the longer you wait, the more interest you pay, so the economic optimum is immediate repayment. The only exception is when the solution is planned to be decommissioned, and the total interest to be paid over the remaining lifespan is lower than the cost of repayment. Usually, however, resources are limited and the business case for technical debt reduction needs to compete with other solution improvements, such as new features.

In all cases, proper representation of the interest is crucial to making a compelling case. In case of structural or technology debt, it is often hard to quantify the extra costs caused by the debt. The difficulty is compounded by the inherent uncertainty: things might go wrong, but they might also go smoothly, even with the technical debt present. One often hears arguments like “we’ve run this application on this platform for 15 years, and it has never caused us any problems – so why invest in an upgrade?”.

The risk factor

The key to making the business case for technical debt reduction is to account for the risk caused by the debt. The proper way to calculate the total expected cost of uncertain failure is the well-known risk exposure formula: E(S) = p(S) x C(S), where p(S) is the probability of failure scenario S occurring, and C(S) is the cost incurred when S occurs. By summing up the risk exposure E over all possible failure scenarios S caused by the technical debt, you come as close as statistically possible to an accurate prediction of the expected cost of failure.

I recently encountered a situation where a large transportation company was running some of their core business systems on ancient mini-computers. Spare parts were very hard to get by, and the manufacturer had put severe limitations on their maintenance contract. The organization in question had a hard time making the business case for migrating the system to a modern, virtualized, blade-based solution: the cost of the old platforms was so low that the ROI for the migration looked negative. The risk of failure, however, was substantial: a single missing spare part could potentially break the company by disabling their core system for a few days. Including that risk exposure in the technical debt interest leads to a completely different business case.

Forgetting the risk factor in the business case for technical debt reduction is a common mistake, which can lead to very wrong decisions. The organization in the example above instinctively knew they had to make the upgrade anyway, but the omission of the risk in the business case did lead to unnecesary delays in the decision making process.

In short: architects arguing for technical debt reduction should make sure that they articulate the risk component in the interest. This will help them convince stakeholders to give proper priority to things they might otherwise find less interesting for being invisible to end-users.

In our architecture classes, we always spend some time discussing the difference between architecture and design. We say that architecture is a set of design decisions, so is there even a difference? In some domains, there really isn’t. In our organization, infrastructure architects generally take all infrastructure design decisions, and are responsible for the completeness of the design they deliver. Software architects sometimes leave detailed design to developers. Solution architects’ projects often encompass multiple technologies and domains, for which they cannot be expected to have mastered all the technology-specific skills to complete the detailed design work. Looking at architecture as a risk and cost management discipline (as RCDA does), the architect should focus on those design decisions that have the highest impact on the risk and cost associated with the solution.

When we ask architecture course attendants (most of whom are practicing architects) for their perception of the difference, a list similar to this one appears on the whiteboard:

The table lists contrasting concepts, of which the one on the left is more generally associated with architecture, and the one on the right with design. The concepts are generally overlapping and not opposing. For example, some design details can have very much impact on the risk and cost of a solution, so an architect can never get away with just worrying about high-level stuff. Sometimes the devil is in the details, so the architect should look at some details – the trick is to know which ones. The fundamental properties come from the ISO 42010 definition of architecture. I think the bottom row is sort of wrong: architects should not communicate with business stakeholders more than with the delivery team, but sometimes it feels that way. The bottom row resonates with the distinction between Architectus Reloadus and Architectus Oryzus in Martin Fowler’s article “Who needs an architect?”

The last few years, I have added two rows to the table:

The first of these is based on how we have seen typical architects and designers deal with uncertainty. When running into an item that has not been decided yet or is otherwise uncertain, successful designers usually put that item on the back-burner. They ask someone (a manager or architect) to resolve the issue, and then quickly move on to analyze the next use case or service item to keep up their productivity. Architects cannot afford to do that. Successful architects actively search for uncertainty, since uncertainty generally is associated with concerns with high impact on risk and cost, that need to be addressed as soon as possible.

The second additional contrast is Conceptual integrity versus Completeness. It pertains to the reason behind having two different roles for architecture and design. What is the main benefit of this role separation? When would you want to put one of the designers of a solution in a different role from the others, and call him architect? This only makes sense if it is too much work for one person to design the complete solution. If you need two or more persons to design a complete solution, it makes sense to appoint one of them to preserve the conceptual integrity across the whole solution. This person we call the architect, and the other designer(s)  should make sure that the solution design is complete. Looking at it that way, the distinction between architecture and design is a separation of concerns. The concerns of conceptual integrity on the one hand and completeness on the other hand need to be separated, because they would otherwise compete for the scarce resource of human attention, at the expense of either one or the other.

So if you are wondering whether your skills are better employed as an architect or as a designer, ask yourself this question: how do you like to deal with uncertainty? At the end of the day, do you get more satisfaction out of a) delivering a nicely rounded, complete design, or b) chasing down unresolved issues and making suboptimal decisions based on incomplete information? If your answer is a), you are probably better off as a designer than as an architect.

Ever since the introduction of “Agile” around the turn of the century, people have wondered how to combine agility with architecture. The difficulty of this combination is due in part to a fundamental difference in goals between the agile and architecture approaches: architects look for stability and future-proof-ness, while agilists want to embrace change, looking for a kind of future-“loose”-ness.

Some say architecture and agility are conflicting approaches. The agile proponents’ attitude towards Big Up-Front Design (BUFD) certainly seems to directly oppose the idea of architecture-driven development. This perception of conflict is increased by the agile movements’ tendency to behave like a religion, complete with dogmas and heretics, as wittily described by Philippe Kruchten. On blogs, agilists sometimes post rants against any suggestion towards up-front thinking about a solution’s architecture, or any hint that not all key (quality) requirements can be addressed afterwards by magically refactoring an IT solution. Fortunately, not all agilists scorn architecture, as witnessed by Scott Ambler’s essay on agile architecture.

Looking more closely, one sees that architecture and agility represent two ends on a spectrum.  Where in this spectrum is the optimal place for your project to live depends on the project context. As noted in an earlier blog post, Barry Boehm suggests that the ideal place on this spectrum depends on three factors that together determine the amount of architecture needed beforehand: the size of the project, the volatility of the environment and the business criticality of the solution.

Agilists can become more successful if they take the project context into account when assessing the usefulness of architecture, but what can architects do to cross the divide between agility and architecture? Judging by TOGAF, the Open Group’s popular architecture framework, the principles in the Agile Manifesto have long been neglected by the architecture community. ADM, the TOGAF architecture development method, still requires rather bulky documentation, produced by often heavy processes like Business Architecture, Information Systems Architecture and Technology Architecture. This type of enterprise architecture approach is not suitable for an agile environment. In the software architecture world, lighter architecture approaches like George Fairbanks’s “Just Enough Architecture” are gradually gaining ground. These more agile approaches no longer view architecting as mainly a design discipline, but also as a way to control risk and deal with uncertainty.

Risk- and Cost Driven Architecture (RCDA) is a relatively new approach in this arena. This approach was developed to close the gap between enterprise and software architecture. Existing software architecture practices are often too limited in scope for the solutions that need to be architected, but the enterprise architecture practices are too heavy for the agility required by time pressure and frequently occurring changes and uncertainty. The RCDA approach incorporates a number of aspects from agile software development practices, such as the use of a backlog of architectural concerns, to be frequently reprioritized based on economic factors like risk and cost.

The secret of making architecting agile is to change your view of the main deliverable of the work, just like with agile software development methods. An agile software development team does not deliver a “big-bang system”, but a continuous stream of improvements to a system. In the same way, an agile architect does not deliver a “big up-front design”, but a continuous stream of architectural decisions, step by step gaining control of the uncertainties and risks surrounding complex IT solutions. How much architecture to build in is determined not by agile dogmas like “You Ain’t Gonna Need It” (YAGNI), but by economic trade-offs taking into account the real value of architecture in context.

Architects can do much to cross the divide towards agility – in fact, they not only can, but must. If they don’t make this effort, architecture departments will lose touch with their IT development departments, where agile methods have become mainstream, and with their business stakeholders, who continually ask for faster and better responses to changing market requirements. The key change architects need to make is to no longer view architecture as a design document to be delivered to projects up-front, but as a continuous decision making process in order to gain control over cost, risks and uncertainty. Only then can architects deliver the added value and flexibility required from them by their business stakeholders.

Listen to Bett and Russ of architecturecast.net interviewing me about these ideas and RCDA.

When reviewing architectures, I sometimes run into architecture description documents that contain long lists of assumptions. There are some negative connotations associated with assumptions. They often read as a “list of things other people should take care of.” They seem to imply that the author:

I recently read an interesting book chapter by Barry Boehm about what is the right amount of architecture to do  “up-front” before committing to the architecture. Boehm analyses his project database and concludes that for most solutions, a proper solution architecture validated up-front will eliminate many of the delivery overruns and shortfalls commonly experienced. The only exception is small, non-critical solutions in a volatile environment; in those situations, architecting generally has a negative ROI.

So what is “small”? Boehm’s analysis yields the following rough estimate of the optimum amount of architecture effort for various project sizes:

My only addition to this table is a translation of Boehm’s Lines of Code to project budget, based on the assumption that the need for architecture in pure software solutions has roughly the same budget relationship as that in IT-related solutions with a wider scope. Boehm also nicely shows that the optimum architecture effort goes down for solutions with a less stable context, and goesup for more business critical solutions.

RCDA on how much up-front architecture

RCDA does not consider architecting to be an “up-front” activity that has to be completed before implementation of the solution can start. In most projects, however, there is a moment that can be identified as the Architecture Milestone. This is the moment after which reversing key architectural decisions becomes very costly and time-consuming, the moment at which the delivery team commits to the architecture. This is a key milestone for solution architects. Solution architects need to know how much architecture needs to be done before this milestone.

In RCDA, the answer to the “how much architecture?” question is based on the view of architecting as a risk- and cost management discipline. Deciding how much architecture needs to be done before committing to the solution is a risk management decision.

Less up-front architecting generally increases the following risks:

• risk of not fulfilling architecturally significant requirements
• risk of rework (refactoring, repairs)

More up-front architecting generally increases other risks:

•  risk of overdesign (YAGNI)
•  risk of idleness (resources waiting to go to work)

As you can see, both options increase risks that can lead to delays in delivery and cost overruns. Which outweighs the other is determined by the solution and project context. The context factors are the same as in Boehm’s analysis:

• A volatile environment increases the probability of overdesign, making less up-front architecting better.
• A highly business-critical solution increases the impact of not fulfilling architecturally significant requirements, making more up-front architecting better.
• A large project increases the impact of both rework and idleness, these two risks compete with each other to determine whether more or less up-front architecting is better. When making the trade-off, take into account that the impact of idleness increases linearly with the project size, while the impact of rework generally increases at a superlinear rate, due to the extra coordination and dependencies between the elements of rework.

In short, RCDA’s view of architecting as a risk- and cost management discipline leads to the same conclusions as Boehm’s analysis, which adds a nice quantitative heuristic for solution architects to use.

References

“Making Software: What Really Works, and Why WeBelieve It“, edited by Andy Oram and Greg Wilson. Chapter 10: “Architecture: how much and when?”, by Barry Boehm.

“RCDA: Architecting as a Risk- and Cost Management Discipline” by Eltjo R. Poort and Hans van Vliet, Journal of Systems and Software (2012)