COE 558Lecture 01Glossary

Glossary

Every term in Introduction, defined once and used the same way in every part. Each entry links to the slides where the idea appears.

Terms: 62
Letters: 20

A

Ad-hoc design: A design made for one immediate situation without prior planning. It can work for small isolated tasks but rarely scales or integrates.
Slide 10
API gateway: A single entry point that receives client requests and routes them to the right backend service.
Slide 34
Slide 35
Application layer: The top layer where network applications and their protocols (HTTP, SMTP, FTP) live. Its unit of data is a message.
Slide 25

B

Backend: The server side of a web application: the server program plus its database, possibly on different machines.
Slide 37
Slide 44
Slide 45

C

Client: The component that initiates communication and presents the user-facing interface, for example a browser.
Slide 14
Slide 15
Slide 39
Cloud: Large, distant data centers with very high compute and storage, organized in regions and availability zones.
Slide 51
Slide 57
Slide 58
Component: A self-contained part of a system, drawn as a box, such as a client, a server or a device.
Slide 14
Continuum: A range of slightly different things between two end points, like the spectrum of light frequencies.
Slide 47

D

Data link layer: The layer that moves frames between two directly connected network interfaces, addressed by MAC addresses.
Slide 25
Slide 26
Slide 27
Decentralized application: An application split into separate client and server programs, usually on different machines, that communicate through a protocol and sockets.
Slide 33
Device: The physical thing being sensed or controlled, such as a lamp driven by a microcontroller.
Slide 14
Slide 17
Distributed application: An application whose computation and state are spread across several machines rather than held in one place.
Slide 33

E

Edge: Compute closest to the client, such as IoT gateways, home routers or a nearby telco site. Low latency, limited resources and reach.
Slide 57
Slide 58
Slide 67
Edge-to-cloud (E2C) continuum: The range of places a service can run, from devices through mist, edge and fog to the cloud, trading latency for processing power and storage.
Slide 46
Slide 56
Slide 57
Encapsulation: Each layer wrapping the unit from the layer above as its payload and adding its own header.
Slide 25

F

Fog: Intermediate nodes or micro data centers between edge and cloud that aggregate, coordinate and partially process data.
Slide 57
Slide 58
Slide 69
Frontend: The client part of a web application, running in the browser with HTML, CSS and JavaScript.
Slide 37
Slide 44
Slide 45

G

Geodesic distance: The shortest path between two points along the curved surface of the Earth.
Slide 52
Slide 53

H

HTTP: HyperText Transfer Protocol, the application-layer request-response protocol of the Web. Methods include GET, PUT, POST and DELETE.
Slide 16
Slide 19
Slide 37

I

Interaction: One exchange between two components, such as a request and its reply. Each interaction is an element of a relationship.
Slide 16
Slide 18
Interface: The operations one component exposes for another to call, such as the endpoints of a Web API.
Slide 16
Slide 19
IP address: The network-layer address that identifies a host interface across the Internet.
Slide 27
Slide 28

L

Latency: The time data spends in transit between two points.
Slide 51
Slide 55
Latency budget: The allowed range [0, τ] for a service's response time, where τ is the maximum acceptable end-to-end response time.
Slide 72
Latency lower bound: The smallest possible one-way latency: geodesic distance divided by the speed of light (about 300,000 km/s).
Slide 52
Layer: A level of a system that groups related functionality and resources, and provides services to the level above it.
Slide 20
Slide 23

M

MAC address: The data-link-layer hardware address of a network interface card, used between directly connected interfaces.
Slide 27
Slide 28
Mapping: An assignment of each task of a pipeline to a layer of the E2C continuum.
Slide 61
Slide 65
Slide 67
Slide 69
Memory hierarchy: Registers, caches, RAM and disk arranged from fastest and smallest to slowest and largest, so frequently used data sits close to the CPU.
Slide 49
Slide 50
Mental model: A simplified internal picture of how a system or process works, used to reason from a problem to a solution.
Slide 8
Mist: The layer on or right beside the device, with very limited resources, used for immediate local decisions.
Slide 57
Slide 58

N

Network application: Any application that uses the Internet as its transport medium, such as the Web, email, Teams or Netflix.
Slide 31
Slide 32
Network layer: The layer that moves packets host to host across networks, addressed by IP addresses.
Slide 25
Slide 26
Slide 27

O

Offloading: Moving work from a weaker component to a stronger one, as a client does when it relies on a server for processing and storage.
Slide 15

P

Physical layer: The layer that transmits raw bits as signals over a wire, fiber or radio.
Slide 25
Port number: A 16-bit number that identifies an application endpoint on a host at the transport layer.
Slide 27
Slide 28
Processing delay: Time a switch, router or host spends examining and forwarding a packet.
Slide 55
Processing time (PT): The time the server spends executing the pipeline tasks for one request.
Slide 62
Propagation delay: Time for a signal to travel the physical length of the channel, set by distance and the medium.
Slide 55
Protocol: The agreed rules and message formats two parties use to communicate, for example HTTP or TCP.
Slide 16
Slide 19
Protocol data unit (PDU): The name of the data unit at a given layer: message, segment, packet, frame or bits.
Slide 25

Q

Queueing delay: Time a packet waits in a buffer at a router or server, driven by congestion.
Slide 55

R

Relationship: The set of all interactions between two components, defined by a protocol plus an interface.
Slide 16
Request-response loop: The client sends a request, the server spends compute, storage and communication to prepare a response, and sends it back. The client always initiates.
Slide 39
Response time (RT): End-to-end time seen by the client: latency to the server plus processing time plus latency back.
Slide 62
Slide 63
Slide 77
Round-trip time (RTT): The time for a complete request and response cycle between two endpoints.
Slide 54

S

Server: The component that hosts services, processes client requests and owns the processing and storage.
Slide 14
Slide 15
Slide 21
Server layer: The layer containing the server software together with the compute, storage and communication that support it. Today it is called the cloud, fog or edge.
Slide 20
Slide 21
Service pipeline: The ordered tasks a service runs to answer one request, for example resize, detect objects, draw bounding boxes.
Slide 61
Socket: A communication endpoint identified by an IP address plus a port number. A connection is identified by the sockets at both ends.
Slide 28
Slide 37
Slide 41
Socket API: The operating system interface between the application layer and the transport layer. A socket is handled like a file descriptor that the program reads and writes.
Slide 29
Slide 41
Structured design: A deliberate process that turns a problem into a system able to handle complexity, scale and integration.
Slide 11
System: An environment in which interactions occur between components.
Slide 13
System thinking: Describing a problem as components, the relationships between them and the layers they sit on, instead of as one opaque thing.
Slide 8
Slide 12

T

TCP: Transmission Control Protocol: connection-oriented, reliable, ordered byte-stream transport, at the cost of connection overhead.
Slide 25
Slide 42
TCP/IP stack: The five-layer model of the Internet: application, transport, network, data link and physical. Connectivity emerges from all five working together.
Slide 22
Slide 23
Slide 24
Transmission delay: Time to push all bits of a packet onto the link, set by packet size and link bandwidth.
Slide 55
Transport layer: The layer that connects one application to another end to end, addressed by port numbers. Its unit of data is a segment (TCP) or datagram (UDP).
Slide 25
Slide 26
Slide 27
Two-dimensional design space: A way to place server components along two axes: decentralization (which layer) and distribution (which compute or storage resource).
Slide 35
Slide 36

U

UDP: User Datagram Protocol: connectionless datagram transport with minimal overhead. Data can be lost or arrive out of order.
Slide 25
Slide 42

W

Web API: An interface a server exposes over HTTP so clients can call its operations, for example a set endpoint reached with PUT and a read endpoint reached with GET.
Slide 19
World Wide Web: The network application of linked resources accessed with HTTP through browsers. It is a subset of what runs on the Internet.
Slide 30
Slide 32