Lower Layers Part 1 Flashcards
What are the three ideal goals when linking network elements?
To exchange data:
1. In any chosen amounts
2. At any chosen speed
3. With zero error rates
Why can’t networks achieve zero error rates and infinite speeds?
Data travels at bounded speeds through networks with finite capacity and experiences non-zero error rates due to real-world constraints.
What metrics are used to characterise a flow of data?
- Bandwidth: Volume of data per unit time (commonly Mbps).
- Latency: Delay in transmission (usually milliseconds).
- Error rate: Errors per gigabit or gigabits per error.
Why is “speed” not always the best measure of network performance?
Speed alone can be misleading, like comparing a car at 60mph to a lorry at 60mph.
It’s more about bandwidth (data capacity) and how much can be transmitted without loss.
How has data transmission capacity evolved over time?
Commodity servers have increased from 10Mbps to 40Gbps in 30 years.
This represents a 4000x increase in bandwidth capacity.
What is latency in networking?
Latency is the time it takes for a bit, or group of bits, to travel through the network.
It includes delays caused by the speed of light, clocking the packet at each end, and delays at intermediate stages due to packet size and bitrate.
Why does the speed of light impose a lower bound on latency?
The speed of light determines the minimum time required for data to travel over long distances, and this is constant. It dominates other delays over long distances.
What is Bit Error Rate (BER), and what causes it?
Bit Error Rate (BER) measures errors in data transmission. Causes include:
1. Data loss due to interference.
2. Cosmic rays, impulse noise, cable, and connector issues.
How does CPU power affect error correction in 2024?
CPU power is cheap, so adding more or longer error detection and correction is rarely costly.
This enables better error handling without significant overhead.
What prediction did Ian make in the slides about memory errors in the next few years?
The prediction is that memory errors in CPU caches will become a significant problem as memory ages and systems handle larger data volumes.
This is especially true as systems process increasingly larger data volumes, requiring caches to handle more intense workloads.
What are the current measurements for measuring volume, latency, and error rates? (long-haul networks)
- Volume: Gigabits or terabits per second.
- Latency: Measured in milliseconds.
- Undetected error rates: 1 bit per terabyte.
Why has latency become a significant issue despite bandwidth improvements?
While bandwidth has increased by 3–5 orders of magnitude (10^3–10^5), latency has not improved much in the last 30 years, making it a major bottleneck today.
What trade-offs do different types of data require? (voice, file transfers and streaming)
- Voice: Tolerates low bandwidth and high errors but cannot tolerate high latency.
- File Transfers (e.g., OS images): Prioritise reliability above all else.
- Streaming and recordings: Tolerate some buffering, unlike live content.
For voice, file transfers and streaming, what are their requirements for latency and error rates?
- Voice and live sport require low latency but tolerate higher error rates.
- Streaming and recordings tolerate latency but require low error rates.
- File transfers and OS images prioritise reliability (low errors) over latency.
Why is latency due to the speed of light significant for long distances?
Because even with infinite bandwidth, latency caused by the speed of light is unavoidable and becomes dominant over long distances, like from London to California.
What is circuit switching, and how did it work in old telephone systems?
Circuit switching connected physical wires so that one microphone was continuously linked to a speaker at the other end using amplifiers and multiplexors.
What are the problems with circuit switching?
- Inefficient: It ties up a duplex circuit even if one or both parties are silent.
- Complexity: Multiplexing was very complicated and expensive, especially with 1950s technology.
- Reliability: Treating the wire as a radio with multiple carriers (for increased efficiency, FDM) was difficult to do reliably.
What is packet switching, and how does it work?
Packet switching divides data into small units called “packets,” adds identifying information, and switches packets over the network to their destination.
What happens when no data is being sent in packet switching?
When no data is being sent, no (or at least few) resources are consumed.
What are the advantages of packet switching over circuit switching?
- Multiplexing happens in the time domain instead of the frequency domain.
- Resources are not wasted when no data is sent.
- You achieve a statistical gain on bandwidth (because it efficiently shares resources among multiple users)
- Data can be re-routed around failed switches, improving resilience.
What is the concept of the Time-Division Multiplexing (TDM) in data streams?
The time domain refers to buffering multiple data streams at a lower rate (e.g., 8kbps) and sending them together at a higher rate (e.g., 16kbps).
The line alternates usage, resulting in latency due to waiting for all bits to arrive.
Does the Time-Division Multiplexing (TDM) approach for data streaming prioritise efficiency or latency?
It prioritises efficiency over latency.
What is the alternative to Time-Division Multiplexing (TDM), and why is it often preferred?
The alternative is frequency domain multiplexing, where data streams are sent simultaneously at different frequencies.
Time domain is often preferred because it is simpler to implement with modern digital electronics and reduces complexity.
Is frequency domain multiplexing faster than time domain?
No, FDM does not make data travel faster.
When is frequency domain better for data streaming?
FDM is better when you have:
- Continuous, independent streams that need to be sent in parallel (e.g., analogue signals like radio stations).
- Sufficient bandwidth to allocate separate frequencies to each stream.
When is time domain better for data streaming?
TDM is better for:
- Digital networks where efficiency matters, and data can be sent in chunks.
- Situations where you can tolerate minor delays while waiting to buffer data (e.g., file transfers).
How does frequency domain multiplexing distinguish signals?
FDM distinguishes signals by their frequency. Each signal is modulated to a different frequency (carrier), allowing multiple signals to share the same medium simultaneously. On the receiving side, filters separate the signals by their frequencies for decoding.
What real-world example illustrates frequency domain multiplexing?
An example is two lighthouses flashing messages in Morse code:
- One uses a red light, and the other uses a green light.
- Filters, like coloured lenses on a telescope, separate the two signals so the messages can be read independently.
How does frequency modulation (FM) work?
Frequency modulation (FM) works by encoding data into a carrier signal by varying its frequency. This allows the data to “ride” on the carrier and be transmitted efficiently over a medium.
Since each signal has its dedicated frequency range and transmits data independently within that range, all signals can flow simultaneously and continuously over the medium. This makes FDM ideal for things like radio and TV broadcasting, where many channels operate at once without interruptions.
What are the key differences between Frequency Domain Multiplexing (FDM) and Time Domain Multiplexing (TDM)?
- FDM does not introduce systematic latency, as signals are transmitted simultaneously on different frequencies.
- TDM introduces latency because data is divided into time slots and sent sequentially, but it efficiently uses bandwidth.
- FDM is harder to engineer robustly and uses less available capacity for real signals.
- TDM is commonly used in networking, whereas FDM is still used in analogue systems like musicians’ monitoring systems. (analogue form (e.g., sound waves or video signals) and are divided by frequency. Unlike digital systems that encode data into discrete 0s and 1s)
What is statistical gain in networking?
Statistical gain refers to the practice where Internet Service Providers (ISPs) oversubscribe bandwidth, assuming that not all users will use their full line rate simultaneously.
For example, selling 10Mbps to 100 users but provisioning only a fraction, such as 5% of the total (contention ratio: 1:20).
This allows efficient resource use but may lead to congestion during peak usage.
What are the main problems with packet-based communication?
- Packets can get lost, delayed, or re-ordered during transmission.
- Data larger than a single packet must be split into packets and reassembled upon arrival.
- Missing packets require waiting for a completed sequence before the data can be used.
What is a virtual circuit in networking?
- A virtual circuit is a connection where the network sets up a path between endpoints.
- The endpoints request the connection, and the network assigns a token to identify it.
- All packets in the connection follow the same route.
- The virtual circuit is torn down when the connection is no longer needed.
What are the key features of virtual circuits?
- The network maintains knowledge of all active connections.
- Packet ordering and loss/duplication are managed but not always guaranteed.
- The user does not need to handle details, although checksums are useful to ensure data integrity.
What is a potential issue with virtual circuits if code crashes?
- If the code managing a virtual circuit crashes, the network may not receive a signal to properly shut down the connection.
- This can leave the virtual circuit active unnecessarily, consuming resources until the network cleans it up.
What is a datagram service?
A service where each packet contains complete addressing information and is treated as a separate item. The network routes packets without needing to know about connections.
What Ian annotation highlights how datagram services send data?
“Datagram services send data across wires and hopes.”
What are Netheads vs Bellheads?
People who mistrust the controllers / government vs people who are closely assigned with the government.
What is the Bellheads’ approach to traffic management?
Bellheads use virtual circuits to shape and groom traffic, adding value with complex protocols.
What is the problem with marking data as priority, as noted in the Ian annotation?
“Everyone would just mark all data as priority.”
What solutions and risks are suggested for preventing all data from being marked as priority?
Solutions: price or flow control.
Risks: government involvement, money scams, or user annoyance.
What problem does layering solve in networking?
Layering allows programs to operate over different types of networks without requiring radical changes.
What is layering?
Layering is the concept of dividing a network into a series of layers or “stacks,” where each layer provides services to the layer above and uses services from the layer below.
How does layering simplify networking for programs?
Layering allows network programs to operate over different types of networks without requiring radical changes, ensuring flexibility and abstraction.
What is an example where layering might not fully apply?
Video streaming might face challenges due to specific network requirements like bandwidth and latency.
What is a solution to enabling flexibility across different network types?
The solution is layering, which provides abstraction and separation of tasks.
