IoT - F 2025

A cyber-physical system consisting of a large number of individually and globally addressable devices which have a significant portion of their logic running elsewhere

Siloed access

Lots of external dependencies

Lack of security

A class of processing regimes which primary property is that it relies on a body of data which expands at a rate where processing has become nontrivial

When going from edge to cloud, processing power increases

When going from cloud to edge, latency decreases

When going from cloud to edge, privacy improves

DSL

Elixir

Erlang

Assembly

Zig

C

Rust

Handle high amount of concurrency

Have a high degree of availability

Offer robust low response times

Be mature

Access to frameworks for modeling information (graphs, ontologies)

Elixir

Erlang

Distribution

Soft real-time properties

Close to zero downtime

Focus on data

Code is an recipe, data is the ingredients

You tell the steps and how it should be executed

The focus is how the result is reached

The focus is on what the program shall achieve

Focus on result

Control flow is not explicitly described

You define what the result is

Declarative programming

Data are immutable

Everything is an expression

Functions are values

Functions are composable through their arguments

Function calls are cheap

There are no loop (in principle)

Functions are first class citizens

Only supports pure functions

No side-effects

No randomness

Functions whose results are depends only on the value of their parameters

When they are idempotent

Two functions that are pure, cannot affect each other

Functions can be executed in parallel

Easy to compile

Easy to test

Hard to write

Pattern matching

Pipelines

Anonymous functions

Higher-Order functions

Returns a tuple, with the first element being an atom indicating success or failure

Returns the result directly, or raises an exception if it fails

Returns a boolean indicating success or failure

Works as a "test"

A process

Consists of state

Consists of Functionality

Consists of a message queue

Actors are isolated from each other

Actors run concurrently

Actors communicate through message passing

Actors are inactive as long as their inbox is empty

Messages are consumed FIFO-style and processed strictly sequentially

Actor is an "Unit of concurrency"

A call

Waits for other process to respond

A cast

Does not wait for other process to respond

A type of process

responsible for administrating the lifecycle of a number processes

Starts and stops processes

Handle crashes of processes

One for one

One for all

Rest for one

Only the crashed process is restarted

Other processes are not affected

If one child process crashes, all other child processes are restarted

The process that crashed is restarted

All process listed after the crashed process are restarted

In a sandboxed process

Handling errors gracefully

Handling errors using if statements

Handling errors using exception handling

Let it fail

It's okay for your code to fail

Having a common interface and interface model between applications and buildings

Persists a values, which are indexed by timestamps

Typically stored as a B+ tree

Used for fast range lookups

Formal naming and definition of types, properties and interrelationships of entities

Exists in a particular domain

Formalized language

Things are defined through the type, parameters and relationships to other things

Limited coverage

Focus on a particular field

Models can span multiple ontologies

The ontology in a model

Envisioning how it will be interpreted

How writing to the model affects model consistency

Think in terms of lower abstractions

Envisioning how it will be interpreted

How writing to the model affects model consistency

Think in terms of lower abstractions

The Resource Description Framework

An ontology defined through triples

Defined through Subject, predicate & object

Namespaces

Subclasses and subproperties

The information lies in the triple

To avoid naming collisions

Realized through compact URI (CURI)

Through prefixes

SparQL

Through their use

Through their place

Removes the ability for any entity to take any role in a triple

Allows modeling the data as a graph

The presumption that any statement that is true is also know to be true

Require all facts to be known

It is possible to validate that all properties hold

The assumption that the truth value of a statement may be true irrespective of wheter or not it is known to be true

Accept that you don't know all facts

A property that needs to be satisfied could reside outside the set of known facts

Impossible to verify that properties hold

Possible to check that no restrictions are violated

Concept

Model

Open World

Classification

Closed World

Validation

Tests the stated rules of the applied ontologies

Spelling mistakes are accepted

Entities can be connected in any way

Entities can be floating

An element that applies rules

When inserting

On triple store before querying

While querying

When all edges have a weight

The level of abstraction refers to nodes, edges, properties and labels

Paths can be matched

A suite of graph algoritms can be referenced

A set of labels

A key-value store containing named properties

A map from relationship type to set of outgoing relationships

A map from relationship type to set of incoming relationships

A single type

A key-value store containing named properties

A source node

A destination node

Garbage collection is problematic in embedded systems

Fast

Low memory footprint

Easy to access low-level operations

Include important high-level abstractions

Simple language

Pre-compilation

Compilation

Linking

Execution

The array may contain garbage values

Variable contains only an address

Contains an integer value

Can be obtained by using the & operator

Can be dereferenced by using the * operator

The difference in electric potential

Measured between two points

Measured in Volts (V)

A device which converts energy from one form to another

A way of transferring a chunk of data one bit after the other

Communication in one direction

Communication in the other direction

Ground

1 start bit

A character spanning 7-8 bits

1 parity bit

Stop bits

Makes it simple to parse

Makes it simple to debug

Lookup using table

Lookup using map

Chain of branches

Deterministic finite automaton

The time between two samples

The number of samples obtained within 1 second

Twice as fast as the fastest frequency contained in the original signal

Nyquist Theorem

Downsampling

Upsampling

Aliasing

Cutting off extreme values

Remove fluctuations

Flush

Wait

Sample

Hold

Compare

The sensor itself

Electromagnetical signals

AD conversion

The inherent noise in a given situation

The minimum signal that can be detected

A measurement that tells us how large a fraction of a signal consists of noise

The signal is doubled

Many types of noise will cancel out

Signal to noise ratio (SNR) goes up

Signal to noise ratio (SNR) is improved

An observation which lies 'far' away from the brunt of distribution

Data about an unexpected event is lost

Information about something thought impossible is lost

A black swan event is lost

9 bits

90 kbps

Frequency

Wavelength

Lower frequency gives longer range

Higher frequency gives shorter range

There exists and open and closed spectrum

The signal extends according to the inverse square law

Media is shared

It is not possible to avoid collisions

The intensity of a signal drops polynomially with distance

It is infeasible to send a signal over a long distance

Bit errors start to occur

Happens when two or more signals overlap in both time and space

High probability of one, some or all signals are lost at the place of collision

Line

Star

Ring

Tiered

Tree

Mesh

Packets are usually made up of individual frames

Frames are ofted dropped due to detected corruption

Frames are often dropped due to complete lack of reception

ACK can be used to acknowledge that a frame has been received

Electrical link

Audio link

Radio link

Reliable delivery of frame to local named receiver

Multi-node routing

Connection-orientation

Multiplexing

Flow control

Session management of connections

Convenient translation

Encryption

Coding

API

Application user

Organizational framework

Rules for organizations and compliance

Large number of incoming data streams

Large number of customers

Similar processing across multiple customers

Expectation of privacy

Small number of processing products presented to the customer

Possibly large number of outgoing data streams

An independent piece of logic

Associated with a state

Associated with zero or more input streams

Associated with zero or more output streams

A node consumes inputs in order to produce outputs

A source, is a node only producing output streams

A sink, is a node only consuming input streams

Flow control

Consumability

Trigger condition

Trigger arbitrarity

Batch size

Fault tolerance

Context switching

Side effects

Networking

Categorization

Validation

Sanity check

If one element is down, it may break the entire pipeline

A communication pattern

Messages can be published through some name on a broker

Clients can subscribe to names on a broker

Whenever a message is published, all clients subscribed to that name are notified

Contains a publisher, broker and client

There exists dumb brokers and smart brokers

There exists dumb clients and smart clients

Kafka is close to dumb broker, smart client

MQTT is in between a dumb broker and a smart broker, and a dumb client and a smart client

"Data Portal" is close to smart broker, dumb client

Kafka subscribe to exact topics

MQTT subscribe using query over topic pattern

sMAP subscribe using query over per-stream key-value store

Due to its large client footprint

Due to its complexity

Due to complexity in topic discovery

Subscribes through topic patterns

Use / as a separator in topics

The + wildcard matches a single level

The # wildcard matches all levels

Supports QoS 0, 1 and 2

Support last will and testament

Support persistent sessions

"home/temperature/#"

"home/temperature/+"

"home/+/sensor1"

"home/+/sensor1/#"

As the number of machines in a system goes towards infinity the time to one of them breaking goes to zero

You have to deal with failures

Built redundancy across systems and machines

BEAM can be used for homogeneous and high-granularity systems

Kubernetes, Ceph, Docker swarm can be used for heterogeneous and low-granularity systems

A collection of independent components that has been combined in a particular configuration in order to solve some problem

Expectations can be formulated as a set of hypothesis

Hypothesis testing is the act of testing to which degree a hypothesis is true

Results of hypothesis testing is usually boolean, floats or probabilistic

Hypothesis testing has its roots in statistics

Hypothesis testing may be done on a subset of a system

They are designed to highlight relevant performance metrics of the phenomenon of interest

They are administered using experimental setup sorrounding the apparatus

They can be done manually or automatically

They need to be repeatable

Examining all combinations of factors

Use all possible parameters for the experiment

An computationally expensive type of test

Include the stubs necessary to execute the system as defined

Operate instances of the relevant components according to fixed values for all parameters

Include stub implementation for all dependencies to the system

Produce the needed workload

Iterate over relevant factors and workloads

Instrument relevant components to observe and log relevant metrics

Not influence the system under experimentation

High-resolution timing is expensive

High-resolution timing is often not critical

Timestamps across machines are often not directly comparable

Timestamps across a single machine may not be directly comparable

Monotonic clocks can be helpful for short periods of time

It is important to understand and document how large the time difference is between multiple systems

Any presented plot of data should be self-contained

Context or context reference should be obvious

Any axis in the plot should be labeled with metric and unit

Ensure axis starts at appropriate zero or known value

Represents the availability of a scarce ressource

Can be consumed

Some get replenished

Can be a defining factor for operation of a device

Number of packets sent throughout a day

Rate of transmission

Maximum packet size

Available size of RAM

Available size of ROM

Available size of flash

Number of per-cell write cycles

Size of erase unit

There are lots of budgets

There are lots of tradeoffs

It is hard to reason about the relative/absolute magnitudes of tradeoffs

It is hard to reason about how design choices affect budgets

There is no one-size-fits-all solution for budget management

You should experiment with different budgets, to identify the best tradeoffs for your system

If one or more values are not sent, the data may not be able to be reconstructed

Blowing the budget, may cause data to not be sent

You will get a sum curve that grows indefinitely

You will transmit the same number of values

You will be able to easily derive the original values from the aggregated values

Data degrades gracefully in temporal resolution

The frequency of sampling is adjusted based on the current state of the system

The frequency is adjusted when changes in the system are detected

Batch transmission can be used to limit number of empty packets

Batch transmission can limit power usage

Batch transmission can increase latency of data

It takes time going from sleep to active state

Low power listening

Receiver listens in short intervals for transmissions

Transmitter waits for receiver to be ready before transmitting

Naming has everything to do with being able to reference something

Identity is the same as a name

A device may have multiple identities

Unique device number

User provided

Physical location

Network-derived

Context-derived

Node - By the device itself (self-aware)

Cloud - Central knowledge base

Private - Stored with the user

When sampling on device

When transmitting

When receiving

When pushing to cloud

When receiving in the cloud

When receiving from the cloud

Originally invencted to communicate bewtween controllers and torpedoes more robust

Signal spread over multiple spectrums

Used in satelite communication

Enchances robustness against interference

Security increased

Less subsceptible to interference

More robust against jamming

Limits detection of communication

Short range

Antenna array that can focus the signal in a specific direction

Focussing done by controlloing signal phases and amplitude across multiple antennas

Utalizes constructive and destructive interference

Improved signal strength

Energie efficient

Reduced interference

Greater range

Faster data rates

Cost

Complexity

A technique to communicate by reflecting existing signals

Reflected signal is modulated in amplitude

Used in RFID

Software-Defined Radio

Allows for programmable hardware

Allows for operating on multiple frequencies and protocols

Zigbee

Bluetooth Low Energy (BLE)

Remote and difficult environments

Disaster recovery and emergency situations

Large events and festivals

Military operations

Smart cities and IoT applications

It's okay that different nodes have different versions of the same data

All nodes will eventually converge to correct version of the data

Quries might return old data

PRoblem: Maintaining data consistency across nodes