While much attention has been focused on high-level software architectural patterns,
what is, in effect, the de-facto standard software architecture is seldom discussed. This
paper examines the most frequently deployed architecture:
the BIG BALL OF MUD.
A BIG BALL OF MUD is a casually, even haphazardly, structured system. Its organization, if one can
call it that, is dictated more by expediency than design. Yet, its enduring popularity
cannot merely be indicative of a general disregard for architecture.
Over the last several years, a number of authors [Garlan & Shaw 1995] [Garlan &
Shaw 1996] [Buschmann et. al. 1996] have presented patterns that characterize high-level
software architectures, such as PIPELINE and LAYERED ARCHITECTURE. In an ideal world,
every system would be an exemplar of one or more such high-level patterns. Yet, this is
not so. The architecture that actually predominates in practice has yet to be discussed:
the BIG BALL OF MUD.
Still, this approach endures and thrives. Why is this architecture so popular? Is it as
bad as it seems, or might it serve as a way-station on the road to more enduring, elegant
artifacts? What forces drive good programmers to build ugly systems? Can we avoid this?
Should we? How can we make such systems better?
These are not anti-patterns, at least in the customary sense. They are not straw men.
Instead, they seek to examine the gap between what we preach and what we practice. They
are set in a context that includes a number of other patterns that we and others have
described. In particular, they are set in contrast to the lifecycle patterns,
PROTOTYPE
PHASE,
EXPANSIONARY PHASE, and
CONSOLIDATION PHASE
, presented in [Foote & Opdyke 1995]
and [Coplien 1995], the
SOFTWARE TECTONICS
pattern in
[Foote & Yoder 1996], and the
framework development patterns in [Roberts & Johnson 1997].
Even systems with well-defined architectures are prone to structural erosion. The
relentless onslaught of changing requirements that any successful system attracts can
gradually undermine its structure. Systems that were once tidy become overgrown as
PIECEMEAL GROWTH gradually allows elements of the system to sprawl in an uncontrolled
fashion.
If such sprawl continues unabated, the structure of the system can become so badly
compromised that it must be abandoned. Like a decaying neighborhood, a downward spiral
ensues. Since the system becomes harder and harder to understand, maintenance becomes more
expensive, and more difficult. Good programmers refuse to work there. Investors withdraw
their capital.
And yet, as with neighborhoods, there are ways to avoid, and even reverse, this sort of
decline. As with anything else in the universe, counteracting entropic forces requires an
investment of energy. Software
gentrification is no exception. A simple way to begin to
control decline is to cordon off the blighted areas, and put an attractive façade around
them. We call this strategy SWEEPING IT UNDER THE RUG. In more advanced cases, there may
be no alternative but to tear everything down and start over. When total RECONSTRUCTION
becomes necessary, all that is left to salvage is the patterns that underlie the
experience.
A major flood, fire, or war may require that a city be evacuated and rebuilt from the
ground up. More often, change takes place a building or block at a time, while the city as
a whole continues to function. Once established, a strategy of
KEEPING IT WORKING
preserves a municipality’s vitality as it grows.
A number of forces can conspire to drive even the most architecturally conscientious
organizations to produce
BIG BALLS OF MUD.
These pervasive, "global" forces are at work in
all the patterns presented.
Among these forces:
One reason that software architectures are so often mediocre is that architecture
frequently takes a back seat to more mundane concerns such as cost, time-to-market, and
programmer skill. Architecture is often seen as a luxury or a frill, or the indulgent
pursuit of lily-gilding compulsives who have no concern for the bottom line. Architecture
is often treated with neglect, and even disdain. While such attitudes are unfortunate,
they are not hard to understand. Architecture is a long-term concern. The concerns above
have to be addressed if a product is not to be stillborn in the marketplace, while the
benefits of good architecture are realized later in the lifecycle, as frameworks mature,
and reusable black-box components emerge [Foote & Opdyke 1995].
Indeed, an immature architecture can be an advantage in a growing system because data
and functionality can migrate to their natural places in the system unencumbered by
artificial architectural constraints. Premature architecture can be more dangerous than
none at all, as unproved architectural hypotheses turn into straightjackets that
discourage evolution and experimentation.
Experience: Even when one has the time and inclination to take architectural
concerns into account, one’s experience, or lack thereof, with the domain can limit
the degree of architectural sophistication that can be brought to a system, particularly
early in its evolution. Some programmers flourish in environments where they can discover
and develop new abstractions, while others are more comfortable in more constrained
environments (for instance, Smalltalk vs.
Visual Basic programmers.) Often, initial
versions of a system are vehicles whereby programmers learn what pieces must be brought
into play to solve a particular problem. Only after these are identified do the
architectural boundaries among parts of the system start to emerge.
Inexperience can take a number of guises. There is absolute, fresh out of school
inexperience. A good architect may lack domain experience, or a domain expert who knows
the code cold may not have architectural experience.
Turnover: Employee turnover can wreak havoc on an organization’s institutional memory, with the
perhaps dubious consolation of bringing fresh blood aboard.
Skill: Programmers differ in their levels of skill, as well as in expertise,
predisposition and temperament. Some programmers have a passion for finding good
abstractions, while some are skilled at navigating the swamps of complex code left to them
by others. Programmers differ tremendously in their degrees of experience with particular
domains, and their capacities for adapting to new ones. Programmers differ in their
language and tool preferences and experience as well.
Complexity: One reason for a muddled architecture is that software often reflects
the inherent complexity of the application domain. This is what
Brooks called
"essential complexity" [Brooks 1995]. In other words, the software is ugly
because the problem is ugly, or at least not well understood. Frequently, the organization
of the system reflects the sprawl and history of the organization that built it (as per
CONWAY’S LAW
[Coplien 1995]) and the compromises that were made along the way.
Renegotiating these relationships is often difficult once the basic boundaries among
system elements are drawn. These relationships can take on the immutable character of
"site" boundaries that Brand [Brand 1994] observed in real cities. Big problems
can arises when the needs of the applications force unrestrained communication across
these boundaries. The system becomes a tangled mess, and what little structure is there
can erode further.
Change: Architecture is a hypothesis about the future that holds that subsequent
change will be confined to that part of the design space encompassed by that architecture.
Of course, the world has a way of mocking our attempts to make such predictions by tossing
us the totally unexpected. A problem we might have been told was definitely ruled out of
consideration for all time may turn out to be dear to the heart of a new client we never
thought we’d have. Such changes may cut directly across the grain of fundamental
architectural decisions made in the light of the certainty that these new contingencies
could never arise. The "right" thing to do might be to redesign the system. The
more likely result is that the architecture of the system will be expediently perturbed to
address the new requirements, with only passing regard for the effect of these radical
changes on the structure of the system.
Cost: Architecture is expensive, especially when a new domain is being explored.
Getting the system right seems like a pointless luxury once the system is limping well
enough to ship. An investment in architecture usually does not pay off immediately.
Indeed, if architectural concerns delay a product’s market entry for too long, then
long-term concerns may be moot. Who benefits from an investment in architecture, and when
is a return on this investment seen? Money spent on a quick-and-dirty project that allows
an immediate entry into the market may be better spent than money spent on elaborate,
speculative architectural fishing expedition. It’s hard to recover the value of your
architectural assets if you’ve long since gone bankrupt.
Programmers with the ability to discern and design quality architectures are reputed to
command a premium. These expenses must be weighed against those of allowing an expensive
system to slip into premature decline and obsolescence. If you think good architecture is
expensive, try bad architecture.
| alias |
| SHANTYTOWN |
| SPAGHETTI CODE |
Shantytowns are squalid, sprawling slums. Everyone seems to agree they are a bad idea,
but forces conspire to promote their emergence anyway. What is it that they are doing
right?
Shantytowns are usually built from common, inexpensive materials and simple tools.
Shantytowns can be built using relatively unskilled labor. Even though the labor force is
"unskilled" in the customary sense, the construction and maintenance of this
sort of housing can be quite labor intensive. There is little specialization. Each housing
unit is constructed and maintained primarily by its inhabitants, and each inhabitant must
be a jack of all the necessary trades. There is little concern for infrastructure, since
infrastructure requires coordination and capital, and specialized resources, equipment,
and skills. There is little overall planning or regulation of growth. Shantytowns emerge
where there is a need for housing, a surplus of unskilled labor, and a dearth of capital
investment. Shantytowns fulfill an immediate, local need for housing by bringing available
resources to bear on the problem. Loftier architectural goals are a luxury that has to
wait.
Maintaining a shantytown is labor-intensive and requires a broad range of skills. One
must be able to improvise repairs with the materials on-hand, and master tasks from roof
repair to ad hoc sanitation. However, there is little of the sort of skilled
specialization that one sees in a mature economy.
All too many of our software systems are, architecturally, little more than
shantytowns. Investment in tools and infrastructure is too often inadequate. Tools are
usually primitive, and infrastructure such as libraries and frameworks, is
undercapitalized. Individual portions of the system grow unchecked, and the lack of
infrastructure and architecture allows problems in one part of the system to erode and
pollute adjacent portions. Deadlines loom like monsoons, and architectural elegance seems
unattainable.
v
v v
As a system nears completion, its actual users may begin to work with it for the first
time. This experience may inspire changes to data formats and the user interface that
undermine architectural decisions that had been thought to be settled. Also, as Brooks
[Brooks 1995] has noted, because software is so flexible, it is often asked to bear the
burden of architectural compromises late in the development cycle of hardware/software
deliverables precisely because of its flexibility.
This phenomenon is not unique to software. Stewart Brand [Brand 1994] has observed that
the period just prior to a building’s initial occupancy can be a stressful period for
both architects and their clients. The money is running out, and the finishing touches are
being put on just those parts of the space that will interact the most with its occupants.
During this period, it can become evident that certain wish-list items are not going to
make it, and that exotic experiments are not going to work. Compromise becomes the
"order of the day".
The time and money to chase perfection are seldom available, nor should they be. To
survive, we must do what it takes to get our software working and out the door on time.
Indeed, if a team completes a project with time to spare, today’s managers are likely
to take that as a sign to provide less time and money or fewer people the next time
around.
| You need to deliver quality software on time, and under
budget. |
Cost: Architecture is a long-term investment. It is easy for the people who are
paying the bills to dismiss it, unless there is some tangible immediate benefit, such a
tax write-off, or unless surplus money and time happens to be available. Such is seldom
the case. More often, the customer needs something working by tomorrow. Often, the people
who control and manage the development process simply do not regard architecture as a
pressing concern.
| Therefore, focus first on features
and functionality, then focus on architecture and performance. |
The case made here resembles Gabriel’s "Worse is Better" arguments
[Gabriel 1993] in a number of respects. Why does so much software, despite the best
intentions and efforts of developers, turn into BIG BALLS OF MUD? Why do slash-and-burn
tactics drive out elegance? Does bad architecture drive out good architecture?
What does this muddy code look like to the programmers in the trenches who must
confront it? Data structures may be haphazardly constructed, or even next to non-existent.
Every shred of important state data may be global. Variable and function names might be
uninformative, or even misleading. Functions themselves may make extensive use of global
variables, as well as long lists of poorly defined parameters. The function themselves are
lengthy and convoluted, and perform several unrelated tasks. Code is duplicated. The flow
of control is hard to understand, and difficult to follow. The programmer’s intent is
next to impossible to discern. The code is simply unreadable, and borders on
indecipherable. The code exhibits the unmistakable signs of patch after patch at the hands
of multiple maintainers, each of whom barely understood the consequences of what he or she
was doing. Did we mention documentation? What documentation?
BIG BALL OF MUD might be thought of as an anti-pattern, since our intention is to show
how passivity in the face of forces that undermine architecture can lead to a quagmire.
However, its undeniable popularity leads to the inexorable conclusion that it is a pattern
in its own right. It is certainly a pervasive, recurring solution to the problem of
producing a working system in the context of software development. It would seem to be the
path of least resistance when one confronts the sorts of forces discussed above.
Only by understanding the logic of its appeal can we channel or counteract the forces
that lead to a BIG BALL OF MUD. One thing that isn’t the answer is rigid,
totalitarian, top-down design. This approach leads to inefficient resources utilization,
analysis paralysis, and design straightjackets and cul-de-sacs.
Kent Beck has observed that the way to build software is to: Make it work. Make it
right. Make it fast [Beck 1997]. "Make it work" means that we should focus on
functionality up-front, and get something running. "Make it right" means that we
should concern ourselves with how to structure the system only after we’ve figured
out the pieces we need to solve the problem in the first place. "Make it fast"
means that we should be concerned about optimizing performance only after we’ve
learned how to solve the problem, and after we’ve discerned an architecture to
elegantly encompass this functionality. Once all this has been done, one can consider how
to make it cheap.
When it comes to software architecture,
form follows function. The distinct
identities of the system’s architectural elements often don’t start to emerge
until after the code is working.
Domain experience is an essential ingredient in any framework design effort. It is hard
to try to follow a front-loaded, top-down design process under the best of circumstances.
Without knowing the architectural demands of the domain, such an attempt is premature, if
not foolhardy. Often, the only way to get domain experience early in the lifecycle is to
hire someone who has worked in a domain before from someone else.
The quality of one’s tools can influence a system’s architecture. If a
system’s architectural goals are inadequately communicated among members of a team,
they will be harder to take into account as the system is designed and constructed.
Finally, engineers will differ in their levels of skill and commitment to architecture.
Sadly, architecture has been undervalued for so long that many engineers regard life with
a BIG BALL OF MUD as normal. Indeed some engineers are particularly skilled at learning to
navigate these quagmires, and guiding others through them. Over time, this symbiosis
between architecture and skills can change the character of the organization itself, as
swamp guides become more valuable than architects. As per
CONWAY’S LAW [Coplien
1995], architects depart in futility, while engineers who have mastered the muddy details
of the system they have built in their images prevail. [Foote & Yoder 1997] went so
far as to observe that inscrutable code might, in fact, have a survival advantage over
good code, by virtue of being difficult to comprehend and change. This advantage can
extend to those programmers who can find their ways around such code.
Yet, a case can be made that the casual, undifferentiated structure of a
BIG BALL OF MUD
is one of its secret advantages, since forces acting between two parts of the system
can be directly addressed without having to worry about undermining the system’s
grander architectural aspirations. These aspirations are modest ones at best in the
typical BIG BALL OF MUD. Indeed, a casual approach to architecture is emblematic of the
early phases of a system’s evolution, as programmers, architects and users learn
their way around the domain [Foote & Opdyke 1995]. During the
PROTOTYPE and
EXPANSIONARY PHASES
of a systems evolution, expedient, white-box inheritance-based code
borrowing, and a relaxed approach to encapsulation are common. Later, as experience with
the system accrues, the grain of the architectural domain becomes discernable, and more
durable black-box components begin to emerge. In other words, it’s okay if the system
looks at first like a BIG BALL OF MUD, at least until you know better.
v
v v
Brian Marick
first suggested the name "BIG BALL OF MUD", and the observation
that this was, perhaps, the dominant architecture currently deployed, during a meeting of the
University of Illinois Patterns Discussion Group
several years ago. We have been using
the term ever since.
BIG BALL OF MUD architectures often emerge from throw-away prototypes, or
THROWAWAY CODE, because the prototype is kept, or the disposable code is never disposed of. (One
might call these "little balls of mud".)
They also can emerge as gradual maintenance and PIECEMEAL GROWTH impinges upon the
structure of a mature system. Once a system is working, a good way to encourage its growth
is to KEEP IT WORKING. One must take care that this gradual process of repair doesn’t
erode its structure, or the result can be a BIG BALL OF MUD.
The
PROTOTYPE PHASE
and
EXPANSION PHASE
patterns in [Foote & Opdyke 1995] both
emphasize that a period of exploration and experimentation is often beneficial before
making enduring architectural commitments.
[Brand 1994] observes that buildings with large spaces punctuated with regular columns
had the paradoxical effect of encouraging the innovative reuse of space precisely because
they constrained the design space. Grandiose flights of architectural fancy
weren’t possible, which
reduced the number of design alternatives that could be put
on the table. Sometimes
FREEDOM FROM CHOICE
[Foote 1988] is what we really want.
| alias |
| QUICK HACK |
| KLEENEX CODE |
| DISPOSABLE CODE |
| SCRIPTING |
| KILLER DEMO |
| PERMANENT PROTOTYPE |
v
v v
A homeowner might erect a temporary storage shed or car port, with every intention of
quickly tearing it down and replacing it with something more permanent. Such structures
have a way of enduring indefinitely. The money expected to replace them might not become
available. Or, once the new structure is constructed, the temptation to continue to use
the old one for "a while" might be hard to resist.
Likewise, when you are prototyping a system, you are not usually concerned with how
elegant or efficient your code is. You know that you will only use it to prove a concept.
Once the prototype is done, the code will be thrown away and written properly. As the time
nears to demonstrate the prototype, the temptation to load it with impressive but utterly
inefficient realizations of the system’s expected eventual functionality can be hard
to resist. Sometimes, this strategy can be a bit too successful. The client, rather than
funding the next phase of the project, may slate the prototype itself for release.
| You need an immediate fix for a small problem, or a
quick prototype or proof of concept. |
Time, or a lack thereof, is frequently the decisive force that
drives programmers to write THROWAWAY CODE. Taking the time to write a proper, well
thought out, well documented program might take more time that is available to solve a
problem, or more time that the problem merits. Often, the programmer will make a frantic
dash to construct a minimally functional program, while all the while promising him or
herself that a better factored, more elegant version will follow thereafter. They may know
full well that building a reusable system will make it easier to solve similar problems in
the future, and that a more polished architecture would result in a system that was easier
to maintain and extend.
Quick-and-dirty coding is often rationalized as being a stopgap measure.
All too often, time is never found for this follow up work. The code languishes, while the
program flourishes.
| Therefore, produce, by any means
available, simple, expedient, disposable code that adequately addresses just the problem
at-hand. |
THROWAWAY code is often written as an alternative to reusing someone
else’s more complex code. When the deadline looms, the certainty that you can produce
a sloppy program that works yourself can outweigh the unknown cost of learning and
mastering someone else’s library or framework.
Programmers are usually not domain experts, especially at first. Use
cases or CRC
cards [Beck & Cunningham 1989] can help them to discover domain objects.
However, nothing beats building a prototype to help a team learn its way around a domain.
v
v v
The production of THROWAWAY CODE is a nearly universal practice. Any software
developer, at any skill or experience level, can be expected to have had at least
occasional first-hand experience with this approach to software development. For example,
in the patterns community, two examples of quick-and-dirty code that have endured are the
PLoP online registration code, and
the Wiki-Wiki Web pages.
The EuroPLoP/PLoP/UP online registration code is, in effect, a distributed web-based
application that runs on four different machines on two continents. Conference information
is maintained on a machine in St. Louis, while registration records are kept on machines
in Illinois and Germany. The system can generate web-based reports of registration
activity, and now even instantaneously maintains an online attendees list. It.began life
in 1995 as a quick-and-dirty collection of HTML, scavenged C demonstration code, and csh
scripts. It was undertaken largely as an experiment in web-based form processing prior to
PLoP ‘95, and, like so many things on the Web, succeeded considerably beyond the
expectations of its authors. Today, it is still essentially the same collection of HTML,
scavenged C demonstration code, and csh scripts. As such, it showcases how quick-and-dirty
code can, when successful, take on a life of its own.
The original C code and scripts probably contained fewer than three dozen original
lines of code. Many lines were cut-and-paste jobs that differ only in the specific text
they generate, or fields that they check.
Here’s an example of one of the scripts that generates the attendance report:
echo "<H2>Registrations: <B>" `ls | wc -l`
"</B></H2>"
echo "<CODE>"
echo "Authors: <B>" `grep 'Author = Yes' * | wc -l`
"</B>"
echo "<BR>"
echo "Non-Authors: <B>" `grep 'Author = No' * | wc -l`
"</B>"
echo "<BR><BR>"
This script is slow and inefficient, particularly as the number of registrations
increases, but not least among its virtues is the fact that it works. Were the
number of attendees to exceed more than around one hundred, this script would start to
perform so badly as to be unusable. However, since hundreds of attendees would exceed the
physical capacity of the conference site, we knew the number of registrations would have
been limited long before the performance of this script became a significant problem. So
while this approach is, in general, a lousy way to address this problem, it is perfectly
satisfactory within the confines of the particular purpose for which the script has ever
actually been used. Such practical constraints are typical of THROWAWAY CODE, and are more
often than not undocumented. For that matter, everything about THROWAWAY CODE is more
often than not undocumented. When documentation exists, it is frequently not current, and
often not accurate.
The Wiki-Web code at
www.c2.com
also started as a CGI experiment undertaken by Ward Cunningham also succeeded
beyond the author’s expectations. The name "wiki" is one of Ward’s
personal jokes, having been taken from a Hawaiian word for "quick" that the
author had seen on an airport van on a vacation in Hawaii. Ward has subsequently used the
name for a number of quick-and-dirty projects. The Wiki Web is unusual in that any
visitor may change anything that anyone else has written indiscriminately. This would seem
like a recipe for vandalism, but in practice, it has worked out well. In light of the
system’s success, the author has subsequently undertaken additional work to polish it
up, but the same quick-and-dirty Perl CGI core remains at the heart of the system.
Both systems might be thought of as being on the verge of graduating from little balls
of mud to BIG BALLS OF MUD. The registration system’s C code
metastasized from
one of the NCSA HTTPD server demos, and still contains zombie code that testifies to this
heritage. At each step, KEEPING IT WORKING is a premiere consideration in deciding whether
to extend or enhance the system. Both systems might be good candidates for RECONSTRUCTION,
were the resources, interest, and audience present to justify such an undertaking. In the
mean time, these systems, which are still sufficiently well suited to the particular tasks
for which they were built, remain in service. Keeping them on the air takes far less
energy than rewriting them. They continue to evolve, in a PIECEMEAL fashion, a little at a
time.
| alias |
| URBAN SPRAWL |
| ITERATIVE-INCREMENTAL DEVELOPMENT |
The Russian Mir Space Station Complex was
designed for reconfiguration and modular
growth. The Core module was launched in 1986, and the Kvant and Kvant-2 modules joined the
complex in 1987 and 1989. The Kristall module was added in 1990. The Spektr and shuttle
Docking modules were added in 1995, the latter surely a
development not anticipated in
1986. The station’s final module, Priroda, was launched in 1996. The common core and
independent maneuvering capabilities of several of the modules have allowed the complex to
be rearranged several times as it has grown.
Urban planning has an uneven history of success. For instance, Washington D.C. was laid
out according to a master plan designed by the
French architect
L’Enfant. The
capitals of
Brazil (Brasilia)
and Nigeria (Abuja) started as paper cities as well. Other
cities, such as Houston, have grown without any overarching plan to guide them. Each
approach has its problems. For instance, the radial street plans in L’Enftant’s
master plan become awkward past a certain distance from the center. The lack of any plan
at all, on the other hand, leads to a patchwork of residential, commercial, and industrial
areas that is dictated by the capricious interaction of local forces such as land
ownership, capital, and zoning. Since concerns such as recreation, shopping close to
homes, and noise/pollution away from homes are not brought directly into the mix, they are
not adequately addressed.
Most cities are more like Houston than Abuja. They may begin as settlements,
subdivisions, docks, or railway stops. Maybe people were drawn by gold, or lumber, access
to transportation, or empty land. As time goes on, certain settlements achieve a critical
mass, and a positive feedback cycle ensues. The city’s success draws tradesmen,
merchants, doctors, and clergymen. The growing population is able to support
infrastructure, governmental institutions, and police protection. These, in turn, draw
more people. Different sections of town develop distinct identities. With few exceptions,
(Salt Lake City comes to mind) the founders of these settlements never stopped to think
that they were founding major cities. Their ambitions were usually more modest, and
immediate.
v
v v
| Master plans are often rigid, misguided and out of
date. Users’ needs change with time. |
Successful software attracts a wider audience, which can, in turn, place a broader
range of requirements on it. These new requirements can run against the grain of the
original design. Nonetheless, they can frequently be addressed, but at the cost of cutting
across the grain of existing architectural assumptions. [Foote 1988] called this
architectural erosion
midlife generality loss.
When designers are faced with a
choice between building something elegant from the ground up, or undermining the
architecture of the existing system to quickly address a problem, architecture usually
loses. Indeed, this is a natural phase in a system’s evolution
[Foote & Opdyke 1995].
This might be thought of as messy kitchen phase, during which pieces of the
system are scattered across the counter, awaiting an eventual cleanup. The danger is that
the clean up is never done. With real kitchens, the board of health will eventually
intervene. With software, alas, there is seldom any corresponding agency to police such
squalor. Uncontrolled growth can ultimately be a malignant force.
In How Buildings Learn, Brand [Brand 1994] observed that what he called High
Road architecture often resulted in buildings that were expensive and difficult to
change, while vernacular, Low Road buildings like bungalows and warehouses were,
paradoxically, much more adaptable. Brand noted that Function melts form, and low
road buildings are more amenable to such change. Similarly, with software, you may be
reluctant to desecrate another programmer’s cathedral. Expedient changes to a low
road system that exhibits no discernable architectural pretensions to begin with are
easier to rationalize.
| Therefore, incrementally address forces
that encourage change and growth. Allow opportunities for growth to be exploited locally,
as they occur. |
In the Oregon Experiment [Brand 1994][Alexander 1988] Alexander noted:
Large-lump development is based on the idea of replacement. Piecemeal Growth
is based on the idea of repair. … Large-lump development is based on the
fallacy that it is possible to build perfect buildings. Piecemeal growth is based on the
healthier and more realistic view that mistakes are inevitable. … Unless money is
available for repairing these mistakes, every building, once built, is condemned to be, to
some extendt unworkable. … Piecemeal growth is based on the assumption that
adaptation between buildings and their users is necessarily a slow and continuous business
which cannot, under any circumstances, be achieve in a single leap.
Alexander has noted that our mortgage and capital expenditure policies make large sums
of money available up front, but do nothing to provide resources for maintenance,
improvement, and evolution [Brand 1994][Alexander 1988]. In the software world, we deploy
our most skilled, experienced people early in the lifecycle. Later on, maintenance is
relegated to junior staff, and resources can be scarce. If the hypothesis that
architectural insight emerges late in the lifecycle is correct, then this practice should
be reconsidered.
Brand went on to observe Maintenance is learning. He distinguishes three
levels of learning in the context of systems. This first is habit, where a system
dutifully serves its function within the parameters for which it was designed. The second
level comes into play when the system must adapt to change. Here, it usually must be
modified, and its capacity to sustain such modification determines it’s degree of
adaptability. The third level is the most interesting: learning to learn. With
buildings, adding a raised floor is an example. Having had to sustain a major upheaval,
the system adapts so that subsequent adaptations will be much less painful.
PIECEMEAL GROWTH can be undertaken in an opportunistic fashion, starting with the
existing, living, breathing system, and working outward, a step at a time, in such a way
as to not undermine the system’s viability. Broad advances on all fronts are avoided.
Instead, change is broken down into small, manageable chunks.
v
v v
A broad consensus that objects emerge from an iterative incremental evolutionary
process has formed in the object-oriented community over the last decade. See for instance
[Booch 1994]. The
SOFTWARE TECTONICS
pattern [Foote & Yoder 1996] examines how systems
can incrementally cope with change.
The biggest risk associated with PIECEMEAL GROWTH is that it will gradually erode the
overall structure of the system, and inexorably turn it into a BIG BALL OF MUD. A strategy
of KEEPING IT WORKING
goes hand in hand with PIECEMEAL GROWTH. Both patterns emphasize
acute, local concerns at the expense of chronic, architectural ones. To counteract these
forces, a permanent commitment to consolidation and refactoring must be
made. It is through such a process that local and global forces are reconciled over time.
This lifecyle perspective has been
dubbed the
fractal model [Foote & Opdyke 1995].
To quote Alexander [Brand 1994][Alexander 1988]:
An organic process of growth and repair must create a gradual sequence of changes,
and these changes must be distributed evenly across all levels of scale. [In developing a
college campus] there must be as much attention to the repair of details—rooms, wings
of buildings, windows, paths—as to the creation of brand new buildings. Only then can
the environment be balanced both as a whole, and in its parts, at every moment in its
history.
| alias |
| VITALITY |
| BABY STEPS |
| FIRST, DO NO HARM |
Probably the greatest factor that keeps us moving forward is that we use the system all
the time, and we keep trying to do new things with it. It is this "living-with"
which drives us to root out failures, to clean up inconsistencies, and which inspires our
occasional innovation.
Daniel H. H. Ingalls [Ingalls 1983]
Once a city establishes its infrastructure, it is imperative that it be kept working.
For example, if the sewers break, and aren’t quickly repaired, the consequences can
escalate from merely unpleasant to genuinely life threatening. People come to expect that
they can rely on their public utilities being available 24 hours per day. They
(rightfully) expect to be able to demand that an outage be treated as an emergency.
Software can be like this. Often a business becomes dependent upon the data driving it.
Businesses have become critically dependent on their software and computing
infrastructures. There are numerous mission critical systems that must be on-the-air
twenty-four hours a day/seven days per week. If these systems go down, inventories can not
be checked, employees can not be paid, aircraft cannot be routed, and so on.
v
v v
| Maintenance needs have accumulated, but an overhaul is
unwise, since you might break the system. |
There may be times where taking a system down for a major overhaul can be justified,
but usually, doing so is fraught with peril. Once the system is brought back up, it is
difficult to tell which from among a large collection of modifications might have caused a
new problem.
| Therefore, do what it takes to maintain
the software and keep it going. Keep it working. |
There may be times where taking a system down for a major overhaul can be justified,
but usually, doing so is fraught with peril. One the system is brought back up, it is
difficult to tell which from among a large collection of modifications might have caused a
new problem.
One of the strengths of this strategy is that modifications that break the system are
rejected immediately. There are always a large number of paths forward from any point in a
system’s evolution, and most of them lead nowhere. By immediately selecting only
those that do not undermine the system’s viability, obvious dead-ends are
avoided.
Of course, this sort of reactive approach, that of kicking the nearest, meanest woolf
from your door, is not necessarily globally optimal. Yet, by eliminating obvious wrong
turns, only more insidiously incorrect paths remain. While these are always harder
to identify and correct, they are, fortunately less numerous than those cases where the
best immediate choice is also the best overall choice as well.
It may seem that this approach only accommodates minor modifications. This is not
necessarily so. Large new subsystems might be constructed off to the side, perhaps by
separate teams, and integrated with the running system in such a way as to minimize
distruption.
Design space might be thought of as a vast, dark, largely unexplored forest. Useful
potential paths through it might be thought of as encompassing working programs. The space
off to the sides of these paths is much larger realm of non-working programs. From any
given point, a few small steps in most directions take you from a working to a non-working
program. From time to time, there are forks in the path, indicating a choice among working
alternatives. In unexplored territory, the prudent strategy is never to stray too far from
the path. Now, if one has a map, a shortcut through the trekless thicket that might save
miles may be evident. Of course, pioneers, by definition, don’t have maps. By taking
small steps in any direction, they know that it is never more than a few steps back to a
working system.
Some years ago, Harlan Mills proposed that any software system should be grown by
incremental development. That is, the system first be made to run, even though it does
nothing useful except call the proper set of dummy subprograms. Then, bit by bit, it is
fleshed out, with the subprograms in turn being developed into actions or calls to empty
stubs in the level below.
…
Nothing is the past decade has to radically changed my own practice, its effectiveness.
…
One always has, at every stage, in the process, a working system. I find that teams can
grow much more complex entities in four months than they can build.
-- From "No Silver Bullet" [Brooks 1995]
Microsoft mandates that a DAILY BUILD of each product be performed at the end of each
working day. Nortel adheres to the slightly less demanding requirement that a working
build be generated at the end of each week [Brooks 1995][Cusumano & Shelby 1995].
Indeed, this approach, and keeping the last working version around, are nearly
universal practices among successful maintenance programmers.
v
v v
Always beginning with a working system helps to encourage PIECEMEAL GROWTH.
SWEEPING IT UNDER THE RUG |
| alias |
| POTEMKIN VILLAGE |
| HOUSECLEANING |
| PRETTY FACE |
| QUARENTINE |
| HIDING IT UNDER THE BED |
One of the most spectacular examples of sweeping a problem under the
rug is the concrete sarcophagus that Soviet engineers constructed to put a
10,000 year lid on the infamous reactor number four at
Chernobyl, in what is now
Ukraine.
If you can’t make a mess go away, at least you can hide it. Urban renewal can
begin by painting murals over graffiti and putting fences around abandoned property.
Children often learn that a single heap in the closet is better than a scattered mess in
the middle of the floor.
There are reasons, other than aesthetic concerns, professional pride, and guilt for
trying to clean up messy code. A deadline may be nearing, and a colleague may want to call
a chunk of your code, if you could only come up with an interface through which it could
be called. If you don’t come up with an easy to understand interface, they’ll
just use someone else’s (perhaps inferior) code. You might be cowering during a
code-review, as your peers trudge through a particularly undistinguished example of your
work. You know that there are good ideas buried in there, but that if you don’t start
to make them more evident, they may be lost.
v
v v
| Overgrown, tangled, haphazard spaghetti code is hard to
comprehend, repair, or extend, and tends to grow even worse if it is not somehow brought
under control. |
There is a limit to how much chaos an individual can tolerate before being overwhelmed.
At first glance, a big BALL OF MUD can inspire terror and despair in the hearts of those
who would try to tame it. The first step on the road to architectural integrity can be to
identify the disordered parts of the system, and isolate them from the rest of it. Once
the problem areas are identified and hemmed in, they can be gentrified using a divide and
conquer strategy.
It should go without saying that comprehensible, attractive, well-engineered code will
be easier to maintain and extend than complicated, convoluted code. However, it takes time
and money to overhaul sloppy code. Still, the cost of allowing it to fester and continue
to decline should not be underestimated.
| Therefore, if you can’t easily make
a mess go away, at least cordon it off. This restricts the disorder to a fixed area, keeps
it out of sight, and can set the stage for additional refactoring. |
By getting the dirt into a single pile beneath the carpet, you at least know where it
is, and can move it around. You’ve still got a pile of dirt on your hands, but it is
localized, and your guests can’t see it.
To begin to get a handle on spaghetti code, find those sections of it that seem less
tightly coupled, and start to draw architectural boundaries there. Separate the global
information into distinct data structures, and enforce communication between these
enclaves using well-defined interfaces. Such steps can be the first ones on the road to
re-establishing the system’s conceptual integrity, and discerning nascent
architectural landmarks.
The UIMX user interface builder for Unix and Motif, and the various Smalltalk GUI
builders both provide a means for programmers to cordon off complexity in this fashion.
v
v v
One frequently constructs a FAÇADE [Gamma et. al. 1995] to put a congenial
"pretty face" on the unpleasantness that is SWEPT UNDER THE RUG. Once these
messy chunks of code have been quarantined, you can expose their functionality using
INTENTION REVEALING SELECTORS [Beck 1997]. This can be the first step on the road to
CONSOLIDATION
too, since one can begin to hem in unregulated growth than may have occurred
during
PROTOTYPING
or
EXPANSION
[Foote & Opdyke 1995].
[Foote & Yoder 1998]
explores how, ironically, inscrutable code can persist because it is difficult to
comprehend. This paper also examines how complexity can be hidden using suitable defaults
(WORKS OUT OF THE BOX
and
PROGRAMMING-BY-DIFFERRENCE),
and interfaces that gradually
reveal additional capabilities as the client grows more sophisticated.
| alias |
| TOTAL REWRITE |
| DEMOLITION |
| THROWAWAY THE FIRST ONE |
| START OVER |
Atlanta’s Fulton County Stadium was built in 1966 to serve as the home of
baseball’s Atlanta Braves, and football’s Atlanta Falcons. In August of 1997,
the stadium was demolished. Two factors contributed to its relatively rapid obsolescence.
One was that the architecture of the original stadium was incapable of accommodating the
addition of the "sky-box" suites that the spreadsheets of ‘90s sporting
economics demanded. No conceivable retrofit could accommodate this requirement. Addressing
it meant starting over, from the ground up. The second was that the stadium’s attempt
to provide a cheap, general solution to the problem of providing a forum for both baseball
and football audiences compromised the needs of both. In only thirty-one years, the
balance among these forces had shifted decidedly. The facility is being replaced by two
new single-purpose stadia.
Might there be lessons for us about unexpected requirements and designing general
components here?
v
v v
| Your code has declined to the point where it is beyond
repair, or even comprehension. |
Obsolescence: Of course, one reason to abandon a system is that it is in fact
technically or economically obsolete. These are distinct situations. A system that is no
longer state-of-the-art may still sell well, while a technically superior system may be
overwhelmed by a more popular competitor for non-technical reasons.
In the realm of concrete and steel, blight is the symptom, and a withdrawal of capital
is the cause. Of course, once this process begins, it can feed on itself. On the other
had, given a steady infusion of resources, buildings can last indefinitely. In Europe,
neighborhoods have flourished for hundreds of years. They have avoided the boom/bust
cycles that characterize some New World cities.
Change: Even though software is a highly malleable medium, like Fulton County
Stadium, new demands can, at times, cut across a system’s architectural assumptions
in such a ways as to make accommodating them next to impossible. In such cases, a total
rewrite might be the only answer.
Cost: Writing-off a system can be traumatic, both to those who have worked on it,
and to those who have paid for it. Software is often treated as an asset by accountants,
and can be an expensive asset at that. Rewriting a system, of course, does not discard its
conceptual design, or its staff’s experience. If it is truly the case that the value
of these assets is in the design experience they embody, then accounting practices must
recognize this.
| Therefore, throw it away it and start
over. |
Sometimes it’s just easier to throw a system away, and start over. Examples
abound. Our shelves are littered with the discarded carcasses of obsolete software and its
documentation. Starting over can be seen as a defeat at the hands of the old code, or a
victory over it.
One reason to start over might be that the previous system was written by people who
are long gone. Doing a rewrite provides new personnel with a way to reestablish contact
between the architecture and the implementation. Sometimes the only way to understand a
system it is to write it yourself. Doing a fresh draft is a way to overcome neglect.
Issues are revisited. A fresh draft adds vigor. You draw back to leap. The quagmire
vanishes. The swamp is drained.
Of course, the new system is not designed in a vacuum. Brook’s famous tar pit is
excavated, and the fossils are examined, to see what they can tell the living. It is
essential that a thorough post-mortem review be done of the old system, to see what it did
well, and why it failed. Bad code can bog down a good design. A good design can isolate
and contain bad code.
Discarding a system dispenses with its implementation, and leaves only its conceptual
design behind. Only the patterns that underlie the system remain, grinning like a Cheshire
cat. It is these that guide the new implementation. With luck, these architectural
insights can be embodied in genuine reusable artifacts in the new system, such as abstract
classes and frameworks. It is by finding these architectural nuggets that the promise of
objects and reuse can finally be fulfilled.
v
v v
The
SOFTWARE TECTONICS
pattern discussed in
[Foote & Yoder 1996] observes that if
incremental change is deferred indefinitely, major upheaval may be the only alternative.
[Foote & Yoder 1998] explores the
WINNING TEAM
phenomenon, whereby otherwise superior
technical solutions are overwhelmed by non-technical
exigencies.
Brooks has eloquently observed that the most dangerous system an architect will ever
design is his or her second system [Brooks 1995]. RECONSTRUCTION provides an opportunity
for this misplaced hubris to exercise itself, so one must keep a wary eye open for it.
In the end, software architecture is about how we distill experience into wisdom, and
disseminate it. We think the patterns herein stand alongside other work regarding software
architecture and evolution that we cited as we went along. Still, we do not consider these
patterns to be anti-patterns. There are good reasons that good programmers build
BIG BALLS OF MUD.
It may well be that the economics of the software world are such that the market
moves so fast that long term architectural ambitions are foolhardy, and that expedient,
slash-and-burn, disposable programming is, in fact, a state-of-the-art strategy. The
success of these approaches, in any case, is undeniable, and seals their pattern-hood.
It is not our purpose to condemn
BIG BALLS OF MUD. Casual architecture is natural
during the early stages of a system’s evolution. The reader must surely suspect,
however, that our hope is that we can aspire to better. By recognizing the forces and
pressures that lead to architectural malaise, and how and when they might be confronted,
we hope to set the stage for the emergence of truly durable artifacts that can put
architects in dominant positions for years to come. The key is to ensure that the system,
its programmers, and, indeed the entire organization, learn about the domain, and
the architectural opportunities looming within it, as the system grows and matures.
A number of people have striven to help us avoid turning this paper into an
unintentional example of its central theme. We are grateful to the members of the
University of Illinois Patterns Group,
John Brant,
Ian Chai,
Ralph Johnson, Lewis Muir,
Dragos Manolescu, Brian Marick, Eiji Nabika, and
Don Roberts who commented on several
particularly raw drafts of this work. We’d like to also thank
our tireless shepherd, Bobby Woolf, who trudged through the muck of several earlier
versions of this paper. Finally, we’d like to acknowledge the members of
our PLoP ’97 Conference Writer’s Workshop. (We'll enumerate them
in full once we get their suggestions incorporated.)
[Alexander 1979]
Christopher Alexander
The Timeless Way of Building
Oxford University Press, Oxford, UK, 1979
http://www.oup-usa.org/
[Alexander et. al 1977]
C. Alexander, S. Ishikawa, and M. Silverstein
A Pattern Language
Oxford University Press, Oxford, UK, 1977
http://www.oup-usa.org/
[Alexander 1988]
Christopher Alexander
The Oregon Experiment
Oxford University Press, Oxford, UK, 1988
http://www.oup-usa.org/
[Beck 1997]
Kent Beck
Smalltalk Best Practice Patterns
Prentice Hall, Upper Saddle River, NJ, 1997
[Beck & Cunningham 1989]
Kent Beck and Ward Cunningham
A Laboratory for Teaching Object-Oriented Thinking
OOPSLA '89 Proceedings
New Orleans, LA
October 1-6 1989, pages 1-6
[Booch 1994]
Grady Booch
Object-Oriented Analysis and Design with Applications
Benjamin/Cummings, Redwood City, CA, 1994
[Brand 1994]
Stewart Brand
How Buildings Learn: What Happens After They're Built
Viking Press, 1994
[Brooks 1995]
Frederick P. Brooks, Jr.
The Mythical Man-Month (Anniversary Edition)
Addison-Wesley, Boston, MA, 1995
[Coplien 1995]
James O. Coplien
A Generative Development-Process Pattern Language
First Conference on Pattern Languages of Programs (PLoP '94)
Monticello, Illinois, August 1994
Pattern Languages of Program Design
edited by James O. Coplien and Douglas C. Schmidt
Addison-Wesley, 1995
[Cusumano & Shelby 1995]
Michael A. Cusumano and Richard W. Shelby
Microsoft Secrets
The Free Press, New York, NY, 1995
[Foote 1988]
Brian Foote
Designing to Facilitate Change with Object-Oriented Frameworks
Masters Thesis, 1988
University of Illinois at Urbana-Champaign
http://www.laputan.org
[Foote & Opdyke 1995]
Brian Foote and William F. Opdyke
Lifecycle and Refactoring Patterns that Support Evolution and Reuse
First Conference on Patterns Languages of Programs (PLoP '94)
Monticello, Illinois, August 1994
Pattern Languages of Program Design
edited by James O. Coplien and Douglas C. Schmidt
Addison-Wesley, 1995
This volume is part of the Addison-Wesley Software Patterns Series.
[Foote & Yoder 1996]
Brian Foote and Joseph W. Yoder
Evolution, Architecture, and Metamorphosis
Second Conference on Patterns Languages of Programs (PLoP '95)
Monticello, Illinois, September 1995
Pattern Languages of Program Design 2
edited by John M. Vlissides, James O. Coplien, and Norman L. Kerth
Addison-Wesley, 1996
This volume is part of the Addison-Wesley Software Patterns Series.
[Foote & Yoder 1998]
Brian Foote and Joseph W. Yoder
The Selfish Class
Third Conference on Patterns Languages of Programs (PLoP '96)
Monticello, Illinois, September 1996
Technical Report #WUCS-97-07, September 1996
Department of Computer Science, Washington University
Pattern Languages of Program Design 3
edited by Robert Martin, Dirk Riehle, and Frank Buschmann
Addison-Wesley, 1998
http://www.laputan.org
This volume is part of the Addison-Wesley Software Patterns Series.
Brian also wrote an introduction for this volume.
[Gabriel 1993]
Richard P. Gabriel
Lisp: Good News Bad News and How to Win Big
http://www.ai.mit.edu/articles/good-news/good-news.html
[Gabriel 1996]
Richard P. Gabriel
Patterns of Software: Tales from the Software Community
Oxford University Press, Oxford, UK, 1996
http://www.oup-usa.org/
[Gamma et al. 1995]
Eric Gamma, Richard Helm, Ralph Johnson, and John Vlissides
Design Patterns: Elements of Reusable Object-Oriented Software
Addison-Wesley Longman, Reading, MA, 1995
[Ingalls 1983]
Daniel H. H. Ingalls
The Evolution of the Smalltalk Virtual Machine
Smalltalk-80: Bits of History, Words of Advice
edited by Glenn Krasner
Addison-Wesley, 1983
[Johnson & Foote 1988]
Ralph E. Johnson and Brian Foote
Designing Reusable Classes
Journal of Object-Oriented Programming
Volume 1, Number 2, June/July 1988
pages 22-35
[Roberts & Johnson 1996]
Don Roberts and Ralph E. Johnson
Evolve Frameworks into Domain-Specific Languages
Third Conference on Pattern Languages of Programs (PLoP '96)
Monticello, Illinois, September 1996
Pattern Languages of Program Design 3
edited by Robert Martin, Dirk Riehle, and Frank Buschmann
Addison-Wesley, 1997
