Lecture #15

Administrivia

• Memorial Day: next Monday

  • no class

• assignment 3

  • published Sunday morning
  • revised Sunday evening
    • thanks to "early adoptors" for beta testing

• assignment 4 lab: cancelled

  • more time to focus on assignment 4
  • distributed application

Previously on CSS 503

• kernel vs. userland
• processes & inter-process communication
• concurrency
• filesystems

• networking/distributed systems

• networking: 2 or more commputers sharing data

• problems created by networking: Fallacies of Distributed Computing

• network topologies

• network of networks: inter-network (internet)

  • switches, routers, gateways (oh, my)

• ethernet packets (frames): metadata (src, dst, ...) + payload

• OSI: 7 layers of protocol

• ethernet/IP/TCP|UDP/application: 4 layers mapped roughly into OSI framework

• each layer provides higher level of abstraction

  • data link: point-to-point communication
  • transport: routing
  • session: reliable connection (circuit)
  • application: high-level protocol (e.g. HTTP)

Internet Protocol (IP) (cont.)

• "best effort" delivery of IP packets

  • unreliable network ("route around damage")
  • packets may be:
    • lost
    • delayed
    • out of order (arrival)
    • duplicate

• TCP & UDP: protocols that sit on top of IP

  • UDP: unreliable, connectionless
  • TCP: reliable connection-oriented
    • reorders out-of-order packets
    • discards duplicates
    • requests retransmit of lost packets

• protocols on top of TCP

  • FTP
  • HTTP

• protocols on top of UDP

  • TFTP
  • audio/video streaming

Hosts & Ports

• IPv4 addresses: 32-bit number (4 octets)
  • text representation: 4 numbers 0..255 separated by period (e.g. 192.168.1.101)
  • running out of addresses (like phone#s)

• IPv6 uses 128-bit addresses (16 bytes)

  • "this time for sure"

• original idea: class A, B, C networks

  • now: netmask
  • CIDR: classless inter-domain routing

• ports

  • TCP/UDP
  • 16-bit numbers
    • service identifier
    • like global number for shared memory
  • /etc/services: text file containing port#s for well-known services
    • 22: ssh
    • 25: smtp
    • 80: http
      • use alternate port# for internal services, e.g. monitoring, REST
        • 8000, 8080, 8888, 8000-9000, ...
    • use distributed lock service (e.g. zookeeper) as "name service"
  • first 1024 ports are "privileged"
    • on Unix-like systems, can only be accessed by root (superuser/admin)

DNS

• IPv4 address: 32-bit number (IPv6: 128-bit number)

  • "convenient" text representation is not very convenient

• database to map symbolic name to IP address

  • "/etc/hosts" file: not very scalable

• Domain Name System

  • interview question (esp. DevOps): how does DNS work (in detail)?

• DNS: "registered" names (name registrar, e.g Go Daddy)

  • very large DB, frequent updates
  • caching for repeated lookups
    • TTL: Time To Live
      • how long to cache
  • hierarchical lookup: foo.bar.com
    • .com is TLD (top-level domain), answers who is responsible for bar.com
      • name server for bar.com return IP address for foo.bar.com