计算机网络（四）

Content Distribution Networks (CDN) and Domain Name System (DNS)

HTTP Overview

• Methods (HTTP/1.1): GET, POST, PUT, DELETE, HEAD.
• Messages:
  • Request: request line, headers, optional body.
  • Response: status line, headers, optional body.
• Statelessness:
  • Pros: scalability, fault tolerance, independent requests.
  • Cons: some apps require state (e.g., shopping carts).
• State Handling:
  • Cookies for client-side state (authentication, tracking).
  • Modern alternatives: cohort-based identifiers (e.g., Google FLoC).

Improving HTTP Performance

• Persistent connections: reuse TCP connections.
• Parallelism: multiple concurrent requests.
• Pipelining: multiple requests over one connection.
• Caching: forward (near clients) vs reverse (near servers).
• Replication: distribute content across multiple servers.

Content Distribution Networks (CDNs)

• Combine caching and replication as a service.
• Benefits:
  • Reduce latency by placing data closer to users.
  • Balance load across servers.
  • Economies of scale.
• Example – Akamai:
  • Rewrites URLs to point to CDN domains (e.g., a128.g.akamai.net).
  • Requests directed to CDN infrastructure.

DNS Overview

• Maps human-readable names to IP addresses.
• Provides indirection: decouples names from addresses.
• History:
  • Initially flat hosts.txt file (not scalable).
  • DNS introduced in 1983 with hierarchy and distributed servers.

DNS Hierarchy

• Namespace: hierarchical (.com, .edu, etc.).
• Administration: delegated zones (e.g., *.umich.edu).
• Server types:
  • Root servers.
  • Top-Level Domain (TLD) servers.
  • Authoritative servers.

DNS Records

• A: hostname → IP.
• NS: domain → authoritative name server.
• CNAME: alias → canonical name.
• MX: domain → mail server.

Name Resolution

• Recursive: server queries on client’s behalf.
• Iterative: server tells client who to ask next.
• Local DNS caching with TTL improves performance.

Reliability and Performance

• Replication of DNS servers for fault tolerance.
• Caching (positive and negative).
• Supports load balancing and CDN redirection.

---

Video Streaming and Cloud Datacenter Applications

Why Video is Special

• Video files are too large to send in one GET.
• Users skip forward or change playback quality.
• Requires adaptive delivery.

Importance

• Dominates Internet traffic (60% in 2020, projected 82% by 2025).
• Major sources: YouTube, Netflix.

The Video Medium

• Sequence of frames (e.g., 30 fps for TV).
• Very large raw size (MBs per second).
• Compression essential.
• Multiple resolutions (480p–4K) for adaptation.

HTTP Streaming

• Client requests video via GET.
• Client buffers before playback to reduce interruptions.
• Issues:
  • Single bitrate → not adaptive to network variability.

DASH (Dynamic Adaptive Streaming over HTTP)

• Client retrieves a manifest file describing available resolutions.
• Requests video in chunks.
• Adapts bitrate based on observed bandwidth.

Cloud Datacenters

• Backbone of modern web services.
• Huge scale: millions of servers, billion-dollar costs.
• Characteristics:
  • Scalability.
  • Fault tolerance with commodity hardware.
  • Multi-tenancy: performance isolation and portability.

Common Applications

• Partition-Aggregate: hierarchical aggregators combine results.
• MapReduce: parallel computing paradigm (e.g., Apache Spark).
• Goal: keep response times <200 ms.

Datacenter Networks

• Traffic types:
  • North-South: client–server.
  • East-West: server–server (dominant in big data).
• Traffic characteristics:
  • Many small flows.
  • Large flows contribute most bytes.

Networking Challenges

• High bandwidth: ideally full bisection bandwidth.
• Oversubscription: core links often bottlenecked.
• Better topologies:
  • Clos topology (multi-stage networks).
  • Provides scalability and high bisection bandwidth.

---

Transport Layer Basics

Purpose of the Transport Layer

• Provides end-to-end communication between processes.
• Services include:
  • Multiplexing/demultiplexing (via ports).
  • Optional reliability, ordering, pacing.

Key Protocols

• UDP: lightweight, only mux/demux, optional checksum.
• TCP: reliable, ordered byte stream, congestion control.
• QUIC: modern transport built on UDP, faster connection setup, encrypted packets.

Sockets and Ports

• Sockets: OS abstraction for communication.
• Ports: 16-bit identifiers.
  • Well-known (0–1023) and ephemeral ports.
  • Used for mux/demux at the host.

Reliable Transport Challenges

• Network may corrupt, delay, drop, reorder, or duplicate packets.
• Solutions require:
  • Checksums.
  • Sequence numbers.
  • Acknowledgements (ACK/NACK).
  • Timers for retransmission.

Protocol Designs

• Stop-and-Wait:
  • Sender sends one packet at a time.
  • Simple but inefficient (throughput limited by RTT).
• Sliding Window:
  • Window of packets in flight.
  • Throughput ≈ min(n × DATA/RTT, link bandwidth).
  • Requires cumulative or selective acknowledgements.

Sliding Window Variants

• Go-Back-N (GBN):
  • Sender retransmits all packets after a lost packet.
  • Receiver discards out-of-order packets.
  • Efficient with low error rates.
• Selective Repeat (SR):
  • Sender retransmits only lost packets.
  • Receiver buffers out-of-order packets.
  • More efficient but complex.

Observations

• Large window sizes can fully utilize links.
• Trade-off between efficiency and complexity:
  • GBN vs SR.
• All reliable protocols rely on: checksums, timers, ACKs, sequence numbers, sliding windows.