Untitled

Buffer Overflow Prevention: Key question is whether we can avoid buffer overflows.
Quality of Service (QoS): Understanding what QoS is and techniques associated with it.
QoS in Internet: Discussion on how QoS is implemented within internet systems.

TCP requires packet loss to identify congestion, typically occurring from buffer overflow.
Alternatives to Avoid Overflows:
- Use Round Trip Time (RTT) to compute average queue lengths - example: TCP Vegas.
- Employ Random Early Detection (RED) to preemptively drop packets before overflow.
- Implement Explicit Congestion Notification (ECN).

Overview:
- Router monitors average queue length to identify congestion.
- Notify end-hosts by randomly dropping packets before queue overflow occurs.
Average Queue Length Drop Probability: A router might establish a drop probability based on queue length.
Active Queue Management (AQM):
- RED exemplifies AQM by proactively dropping or marking packets to let sending sources reduce their rates before buffers are full.

ECN seeks to inform end-hosts about congestion rather than allowing congestion detection to be passive.
Potential Approaches:
- Congested router returns a congestion notification directly to the source.
- Modifying a congestion-indication bit within packets that allows the destination to inform the source of congestion.
- Create special packets sent by the source to assess congestion levels at routers.
For the internet, the second approach (modifying bits) has been adopted.

Two bits in the IP header denote congestion status:
- 00: transport layer not ECN-capable.
- 01: transport layer ECN-capable.
- 10: transport layer ECN-capable, but congestion NOT experienced.
- 11: transport layer ECN-capable and congestion experienced.
Router Behavior:
- Uncongested routers retain the bits unchanged. Congested routers drop packets if transport is non-ECN-capable; otherwise, they set bits to 11.

Routers signal congestion in packets traveling from A to B and A must adjust the sending rate accordingly.
The transport layer should echo back the congestion notification to sender A using a congestion-experienced notification bit (ECE).
Signal Flow: A sends packets to B, with ECN bits being updated as necessary.

Real-time applications require assurances such as “deliver on time”.
Example audio application:
- A sampler and A/D converter generate one byte 8000 times per second.
- Samples must be played back 8000 times per second.
- Variables include potential network delays and the risk of packet loss.
Current IP and UDP Limitations:
- IP (UDP) offers no delivery or delay guarantees, only best-effort guarantees.
- TCP provides delivery guarantees but can introduce significant delays through retransmissions, making it unsuitable for real-time applications.
Many audio/video applications use TCP even though they face delays from retransmissions, particularly where delay is not critical (e.g., listening to a radio station).

Scarce Resources: Questions arise regarding which resources (like link capacity and buffer space) to allocate and how to schedule packets effectively.
Decisions in Network:
1. Scheduling Discipline: Determines packet service order based on resource availability.
2. Drop Policy: Specifies packets’ queue treatment when resources are limited.

First In, First Out (FIFO): No prioritization; all packets treated equally.
Priority Queueing: Classifies packets into priority groups; higher priority packets served first.
Round Robin Scheduling: Handles packets by class per round; may not be fair due to varying packet sizes.
Bit-by-Bit Round Robin: A theoretical approach where one bit is served from each class per round; not feasible in practice due to complexities.

Tries to mimic bit-by-bit round robin scheduling, ordering packets so they leave the queue as if in bit-by-bit order without interrupting those already being sent.
Weighted Fair Queueing: Extends fair queueing by assigning weights, allowing for uneven bandwidth distribution between flows.

Effective QoS relies on several components:
- Queueing Discipline
- Packet Classification: Determines which packet receives which treatment.
- Admission Control: Ensures network capacity for new and existing flows.
- Traffic Policing: Verifies that flows adhere to their defined traffic specifications.

Describes traffic characteristics using two parameters:
- r: maximum average bytes per second.
- B: maximum burst size in bytes.
Token Bucket Behavior: If a data stream adheres to the token bucket model, it guarantees average sending rates within specified bounds while allowing bursts up to parameter B.

Three Approaches to QoS:
1. Integrated Services (IntServ): Reservations made for individual flows at routers.
2. Differentiated Services (DiffServ): Groups flows into classes, focusing on class-level service rather than per-flow.
3. Overprovisioning: Simply providing excessive bandwidth as a solution.

Enables link capacity reservations for flows needing it, using protocols like RSVP.
Enforces admission controls to sustain quality guarantees.
Provides two service categories:
- Guaranteed Service: Delay assurances for flows.
- Controlled Load Service: Network experience similar to lightly loaded scenarios.
Challenges: Significant scalability issues due to state management and classification.

Limits the number of traffic classes defined and their expected service.
Classification at network edges, while interior routers manage routing based on marked classes.
PHB example: Expedited Forwarding (EF) guarantees serving rates and low delays, frequently used in VoIP applications.

Importance of determining which resources to allocate to which packets.
Congestion Control: Essential to prevent service degradation by adjusting loads based on congestion signals.
Routers must enforce queuing policies to dictate packet treatment.
Early signaling of congestion through strategies like RED and ECN.
QoS: Aims for service differentiation through careful classification and admission controls, though it’s not universally deployed in the Internet.