Evolutionary architecture: What can we learn from Nature?

Evolutionary Architecture: What can we learn
from Nature?
Chris Howe-Jones
4th April 2019
Who am I?
Name: Chris Howe-Jones [Technical Navigator]
Twitter: @agile_geek
Github: https://github.com/chrishowejones
• Been programming for over 40 years, professionally for over 30.
• I’ve been designing software systems for over 25 years
• Started in monolithic homogeneous environment - mainframes
• Automating basic processes in a large government org that changed
rules every couple of years with change of the minister but within
parameters mainly controlled by the speed with which the org could
respond.
1

• Environment was constrained and could only change at a fairly slow
rate
• In the 90’s/00’s software centred companies competed with new fea-
tures as their selling point so speed of change became a competitive
advantage.
• Now we work in markets driven by customers that interact directly
rather than indirectly.
• Companies are built up around providing services that don’t exist be-
fore the company introduces it.
– Facebook (although arguably MySpace was also in that field)
• Companies that are build a business model around a uniqueness -
Google
• Companies even more frequently provide solutions to problems that
already have solutions but they add some additional novelty, usually
in the form of features that provide a different form of interaction with
the customer that didn’t exist before the software was written.
– Apple does this but often in the form of physical design combined
with software e.g. the iPod wheel as the hook and iTunes which
provided an interface that other MP3 download sites didn’t have.
– Slack does what IRC has done for 30 years but provided a UI
that was more tactile and visual (introducing previews for images,
emoji) but also, the killer feature, was a persistent history making
the communication more async as you don’t need to be online at
the same time. Also providing a fermium model and removing
the hosting barriers of some competitors. Having the message
limits meant you can be away from the convo for a while but
not too long introducing a compulsion to check regularly. Also
focusing on teams rather than threaded comments, initially they
implemented everything as a continuous stream which encouraged
constant checking which is encourages compulsive behaviour.
• We see companies starting with one vision of their offering but changing
direction, sometimes quite radically.
– Stuart Butterfield has done this a few times:
2

∗ Ludicorp founded 2002 to build a game (called Game Nev-
erending)
∗ 2004 - game failed to gain traction but embedded photo-
uploading was popular.
∗ 2005 - game crumbled but Flickr rose from ashes
∗ 2006 - Flickr sold to Yahoo
∗ 2008 - Stuart left Yahoo to found Tiny Spec creating a game
called Glitch.
∗ 2010 - raised Series A funding, no detail about game but
it was described as web-based massively-multiplayer game
(Massively Mulitplayer Online Role-Playing Game MMORPG)
epected launch 2010.
∗ 2011 - raised more funding. Invite-based Beta. Launch but
returned to Beta to make it easier for new players
∗ 2012 - became obvious Glitch could not attract audience large
enough to sustain itself. Glitch dead- Tiny Spec was not.
∗ 2013 - internal comms tool used to share ideas between the
US and Canadian oﬃces of Glitch recognised as valuable
∗ 2014 - Slack was announced.
3

• Seeking different analogies - engineering is not a great one are there
other perspectives?
– Civil Engineering - Building the Tyne Bridge. No one widens the
Tyne by 300 miles and relocates it to the South Pole mid project.
– Engineering in Aerospace - flying a plane or guiding a spacecraft
is defined by simple Newtonian physics. Lots of variables but
simple equations.
– What about biology?
4

Mapping Software to Biology
• How would software design map to biological systems?
• How do classes/functions map to cells?
• Cells are self contained - but communication or cooperate to provide
higher functions
5

Mapping Software to Biology
• Message passing like neurochemical transmission
– diﬀerent types:
∗ neurotransmitters - message calling between methods/functions?
(chemical synapse - such as gultamate)
∗ blood - async message based systems, stream processing
∗ nerve endings/hormones - endpoints recieving stimuli in the
form of ’data’ from the outside world.
6

Messages
7

Messages
8

S/w Environments -> Ecosystems
• Ecosystem in nature
• At various levels
• Complex system subject to changes in temp, moisture, destructive
forces.
9

S/w Environments -> Ecosystems
• Ecosystem at various levels
– One of the issues is what state is the Ecosystem in?
– This is not obvious, how do we measure or detect what the Ecosys-
tem is doing as a whole while it’s active?
– How are parts of the ecosystem reacting in coordination to others?
• Changes to the Ecosystem are changes to features, requirements, etc.
Large changes are changes in climate, diﬀerent ecosystem, etc.
• Moving a specialised organism to a completely diﬀerent ecosystem
means it will struggle to survive or may simply die immediately.
• Evolution is the organism changing over time to better exploit its
ecosystem - the mechanism of evolution is the development team.
10

DDD Bounded Contexts
• Are DDD bounded contexts the equivalent of separate organisms?
• They evolve at there own rates. They may be aﬀected by the same
stimuli but they don’t share internal structures.
• Abstractions within software (interfaces and data structures) are anal-
ogous to membranes and sensory mechanisms in organisms.
11

Monoliths
Monoliths are like a single organism and the ’bounded context’ are the
organs . Organs can share nutrients, communication mechanisms (ganglia,
neurotransmitters).
12

Monoliths
Packages or namespaces are bounded context equivalent to organs Comms
via in process calls
13

Services
Services are individual organisms the boundaries of which are the ’bounded
context’. Each service is separate. They evolve at their own rates. They may
be aﬀected by the same stimuli but they don’t share internal structures.
14

Microservices
Microservices are like a Siphonophorae, like a Portuguese man o’ war.
A bounded context is a colonial organism made up of small organisms that
cooperate to get a task done. There are diﬀerent types of organism with
specialised functions zooids or polyps - dactylozooid (defense), gonozooid
(reproduction), and gastrozooid (feeding)
Problems
Negative Environmental Factors -> Design Faults
15

Problems
Poor architecture in:
• Monoliths - cancer(uncontrolled growth), disease (breakdown of shared
classes, namespaces, etc).
• Stress factors in cells (hormones) cause DNA knots that interfere with
RNA transcription.
16

Problems
Poor architecture in:
• Services - environmental changes - polution, poisonous environmental
factors, extremes of heat and cold. Each individual service is subject
to disease, cancer, etc.
• Microservices - less impacted by disease, cancer as they’re smaller/simpler
organisms less to go wrong. Environmental changes in the colony eﬀect
the whole colony. We can lose an individual in the organism type in
the colony and continue.
Individual Organisms mechanisms
• So if each organism is a monolith, a service or a microservice that
operates within the bounds of it’s ecosystem.
• How do they sense and communicate with the ecosystem or other ’or-
ganisms’?
17

-ve feedback mechanisms
• Damp down the effect of a stimulus
• Homeostatic control (negative feedback)
1. Stimulus– produces a change to a variable (the factor being reg-
ulated).
2. Receptor– detects the change. The receptor monitors the envi-
ronment and responds to change (stimuli).
3. Input– information travels along the (afferent) pathway to the
control center. The control center determines the appropriate
response and course of action.
4. Output– information sent from the control center travels down
the (efferent) pathway to the effector.
5. Response– a response from the effector balances out the original
stimulus to maintain homeostasis.
• Temperature regulation in primates
1. Sensors detect rise in temperature.
18

2. Nerve pathways send messages to the control centres in the brain
(hypothalamus)
3. Hypothalamus sends signals to sweat glands to produce sweat.
4. Skin cools
-ve feedback mechanisms
Circuit Breaker on a client.
1. Clients send requests to many requests for a service to handle.
2. Service starts to reject requests.
3. Client implements a circuit breaker to back oﬀ until service recovers
or another service is spun up to take on requests.
19

+ve feedback mechanisms
A positive feedback loop comes into play during blood clotting.
1. Once a vessel is damaged,
2. platelets start to cling to the injured site and release chemicals
3. that attract more platelets.
The platelets continue to pile up and release chemicals until a clot is
formed.
20

+ve feedback mechanisms
Positive feedback in software is driven by external agencies such as cus-
tomer demand, management demand, changes in organisational focus.
Studies have show that too much positive feedback promoting large changes
in a single generation (version) of software systems tends to have a destabil-
ising effect if there is not a commensurate negative feedback loop to dampen
the effect. This can be seen in the 8 Laws of Software Evolution documented
by M.M.Lehman.
Positive feedback loops in software development (evolution) are the intro-
duction of new features, new capabilities, changes in business model resulting
in massive changes in structure and capabilities of not just the software de-
sign itself but in the structure of the team and the processes adopted by the
team in software development (evolution).
This is not always negative - positive feedback introducing changes in
development process or materials can provide balanced growth if the changes
introduced have inherent constraints that provide a negative feedback loop
or dampening effect to appropriate characteristics of the system design or
the development process.
As an example, introducing a new programming language that has fea-
tures that encourage constraints such as immutable data and pure functions
21

can reduce the complexity of concurrent processing and simplify the develop-
ers mental model to not have to consider how state changes over time or race
conditions etc. except at the edges of the system/component/service where
state changes are pushed i.e. HTTP request/response, I/O to ﬁle systems or
databases.
Another example of positive feedback in the form of a change to the
development process that intentionally introduces a constraint to guide de-
velopment might be the introduction of WIP limits on a Kanban board.
WIP limits ensure the development team are only working at the capacity
they can sustain. They reduce the amount of waste in the process in the
form of work at rest (waiting to be started). They also highlight when the
team has too much work concurrently to manage. WIP limits also make
it easier to tune cycle time (time between the work starting and delivery/)
and lead time (time between a feature/story/requirement being raised and
delivered).
Ecosystems
• Diﬀerent levels of Ecosystem
22

Micro-ecosystem
In biology an example of a microecosystem might be a pond which has a
limited amount of variation. It’s temperature diﬀerences are buﬀered. Water
levels can raise or fall but are usually in predictable bounds.
23

Micro-ecosystem
1. Software executables
2. OS
3. Runtime environments
4. Monitoring and logging.
5. etc.
24

Habitat or Biotope
In biology an example of a habitat might be the garden the pond is in.
It has a limited number of species of ﬂora and fauna. Predictable amounts
of rainfall, temperature, shade etc.
25

Habitat or Biotope
1. Development processes
2. Development team interactions
3. Software languages
4. Software tools
26

Biome
In biology an example of a Biome would be Tropical Rainforest, Tundra,
Hot Desert, etc.
27

Biome
1. Department or organisation.
2. Customer base
3. Target market
4. Competitors in the same market
28

Biosphere
In biology a biosphere is planetary scale e.g. Earth.
29

Biosphere
1. All markets
2. All potential customers
3. All competitors
Evolutionary Mechanism
• Adaption
• Acclimatisation
• Other mechanisms
Adaption
• Mechanisms that involve large changes to respond to environment pres-
sures in the ecosystem. For example, a large and permanent tempera-
ture change, the introduction on competitive species.
30

• Adaption involves changes in the phenotype, therefore involving a new
generation (version) of the ’species’. For example, adult stature or eye
colour.
Acclimatisation
• Mechanisms that involve typically smaller variations in phenotypically
plastic characteristics. For example, adjustments in heart rate, skin
colour and attention span.
Other mechanisms
• there are mechanisms in Epigenetics where external or environmental
factors result in a heritable phenotype change or phenotypically plastic
changes. For example, stress factors in the environment may trigger
a gene to activate or inhibit the expression of a gene without actually
changing the nucleotide sequence.
Changes in Biome/Biotope
Biological responses
31

• Most organisms can’t respond to major changes in their Biome by
acclimatisation as the small variations possible are usually changes
that are only capable of responding to the smaller variations that are
conﬁned to the local ecosystem of the habitat or biotope.
• Gradual or smaller changes to the biome can be responded to by adap-
tion that happen in one or two generations through epigenetic mech-
anisms such as phenotype plasticity. These might be something like
a dormant gene being activated by external factors. Morphological
plasticity can be seen in pond snails that develop spiny shells in the
presence of predators.
• Larger changes in the biome require more major genetic changes in
the DNA and occur through a number of mechanisms, the best un-
derstood of which is gene mutation but recent research suggests other
mechanisms such as stress hormones altering RNA transcription and
the adoption of phenotypically plastic traits.
Changes in Biome/Biotope
Software response
32

• The possible response mechanisms in software are controlled by the
development team and associated stakeholders. The team can adopt
techniques that provided a ’guided’ evolutionary mechanisms.
• Smaller changes can be made in days in one generation (version) either
by tweaking parameterised features or making small code changes to
the internal mechanisms in a service or application. Generally speaking
making these changes quickly is much easier if the software component
is smaller.
• Larger changes to response take more major changes internally but also
tend to change the interfaces of the software component. I.e. the inter-
face for clients or the persistent data used is changed. These changes
tend to alter factors that a coupled to other software components or
to external customers/systems.
• What is frequently overlooked is that often this kind of change has an
impact on the more complex microecosystem. As an example, making
a change to add something to an API changes the responsibilities of
the software component involved in a way that may mean that the
component is now taking on responsibilities that it didn’t have before.
This can result in other software components at a later time connecting
to the first to use this new responsibility. This kind of network effect
increase the amount of communication and coordination required and
hence may increase cross network traffic and result in hot-spots in the
architecture.
Feedback loops
Feedback loops in Biology
• Smaller variations - behavioural changes, physiological changes, phe-
notypically plastic traits
• Genetic mutation, cooperative strategies.
• Different species move into ecosystem, most drastically extinction.
33

Feedback loops
Feedback loops in Code
• In general mechanisms to respond to stimuli that are built into ’code’
or even ’conﬁguration’ tend to be negative feedback loops that respond
to a an external stimuli by damping down or stabilising the software
system.
• Good examples of this would be circuit breakers, bulkhead patterns or
rate limiting.
Feedback loops in Development team and process
• The development team and the processes and tools that they use are
a mix of negative feedback loops and positive feedback loops. Gen-
erally speaking, these changes are adopted to facilitate faster changes
in certain directions. However, conversely, often the approaches and
tools are successful in this because they have constraints (or negative
feedback) that provides guide rails that push development towards the
desired characteristics.
• As examples:
34

1. adopting Kanban properly can greatly increase the speed of re-
sponse to changes in the habitat but Kanban as a process has at
it’s heart strong constraints on the amount of work in progress at
any point in time.
2. adopting a functional programming language with immutable data
structures can reduce the effort involved in developing software
that has concurrent processes.
3. adopting a microservices architecture that constrains the respon-
sibility of a service to just one small part of a domain can result
in the ability to respond faster to changes as a microservice can
be changed in hours or replaced in days or a week.
Guided Evolution
Development Team are the primary evolutionary mechanism
In order to guide software evolutionary processes the team needs to un-
derstand the ecosystem they are designing for. Traditionally, this has been
an exercise in gathering and fixing requirements/features/stories but this
is often flawed as to accurately determine what the Biotope is using static
analysis based on peoples interpretation is inherently unconsciously biased.
In order to determine what the ecosystem is we need to ’sense’ the envi-
ronment we are in.
Sensing micro-ecosystem
• Traditional monitoring of running software. Memory, CPU, etc.
• Monitoring and alerting on network effects like request rates, aver-
age/std deviation of request latency, etc.
Sensing the Habitat/Biotope
• Dwell time on a site
• Conversion rates
• Sales rates
• Return rates
35

• Usage rates of features
• Customer location
• Customer segmentation
• A/B testing
• Scientist type code experiments. Code that performs old and new way,
compares results and reports diﬀerences while taking the old result as
response.
Sensing the Biome
• Surveying current customers and wider customer base
• Market research
• Industry research
• Exploratory features or new apps targeting other markets
• P & L
• Board level decisions
Sensing the Biosphere
• Major organisational events (mergers, buyouts, bankruptcy proceed-
ings)
Feedback from sensors
• tune evolution
• Feedback from these sensors are used to tune the evolutionary processes
at the appropriate level.
• You can move to respond to changes in a wider ecosystem by changing
the processes and the sensors to guide evolution to respond to the large
change over generations of the software design.
• Conversely, detecting changes in a sensors output over time in a nar-
rower ecosystem can suggest that a large change is occurring in a wider
ecosystem.
36

Factors that influence software evolution
• Not designing a sensor/response feedback mechanism
• Bad development practices making response to change harder or im-
possible
– Poor code hygene
• Constraints on the ecosystems that inhibit the ability to sense or re-
spond to change.
– Typically, edicts to follow a directive/std with no understanding
or explanation of the reasoning behind this.
• Over population effects poisoning the ecosystem.
Conclusion
• Identify the factors that matter and pick metrics that measure them
• Design your feedback loops (sensor/receptors) and effectors
• Identify beneficial constraints vs detrimental constraints
– e.g. service boundaries or immutable data vs inability to change
biotope due to policy/stds
– Ensure Ecosystems are self contained
– Constrain & understand changes from upper ecosystem.
• People (the team) are your mechanism for evolution - they need to
know what constraints are imposed and why and the directions of
freedom as well as the results of sensory input.
Citations
• Positive feedback and alternative stable states in inbreeding, coopera-
tion, sex roles and other evolutionary processes - Jussi Lehtonen and
Hanna Kokko
• Epigenetic Feedback Regulation Accelerates Adaptation and Evolution
- Chikara Furusawa, Kunihiko Kaneko
37

• A feedback model of evolutionary adaptation - Daniele Casagrande,
Wiesław Krajewski, Umberto Viaro
• Differential Evolution - Clever Algorithms: Nature-Inspired Program-
ming Recipes - Jason Brownlee PhD
• On the nature of change - Brendan Montague
• Listening to nature: How sound can help us understand environmental
change - Garth Paine
• How can a Biological System be affected by a change in pH levels -
JacobS
• Fold-change detection in biological systems - MiriAdler, UriAlon
• On the relation between fluctuation & response in biological systems -
Katsuhiko Sato, Yoichiro Ito, Tetsuya Yomo, and Kunihiko Kaneko
• DNA and Knot Theory - M. Beals, L. Gross, S. Harrell
• 6 Animals that are Rapidly Evolving - Jessica Hullinger
• 5 Examples of How Organisms respond to Stimuli - Austin Brashear
Brashear
• Cells can respond to changes in their environment - Berg JM, Ty-
moczko JL, Stryer L
• A Guide to Neurotransmitter Balance - https://poweronpoweroff.
com
• Difference between Adaptation and Acclimation - maureen. Difference-
Between.net. March 15, 2011
• Animal Sensory Systems - Shana Kerr. November 13, 2016
• Microservices, biology and evolution - Sajid Kalla. medium.com
• The role of behaviour in evolution - Prof Adrian Lister, Prof Ian Barnes.
Natural History Museum London
• How do ants assess food volume? - Anne-Catherine Mailleux, Jean-
Louis Deneubourg & Claire Detrain
38

• Laws of Software Evolution Revisited - M.M.Lehman
• Building Evolutionary Architectures - Neal Ford, Rebecca Parsons &
Patrick Kua
• Domain Driven Design Distilled - Vaughn Vernon
39

Evolutionary architecture: What can we learn from Nature?

More Related Content

Similar to Evolutionary architecture: What can we learn from Nature? (20)

Recently uploaded (20)

Evolutionary architecture: What can we learn from Nature?