(4) Network History

• First regarded as a convenient workaround for floppy disks
  • Real computer scientists write compilers
  • The value of computer networks depends on their size
• Early networking solutions were vendor-specific islands
  • DECnet for Digital Equipment Corporation (DEC) customers
  • XNS for Xerox customers
  • SNA for IBM customers
  • transmitting data between these networks was very cumbersome
• Bridging networks transparently became increasingly important
  • More computers and networks increase the benefit of interconnections
  • Layering being used for internetworks, not only for networks

(5) Networks vs. Internetworks

• Specific networks use specific abstractions
  • How to address nodes (computers, phones, PDAs, RFID tags)
  • How to address applications on these nodes
  • How to transmit data to these applications
• Internetworks provide a network-independent abstraction
  • Nodes are addressed uniformly (IP addresses)
  • Applications are identified uniformly (ports)
  • Data transmission uses one set of protocols (TCP/UDP)

(6) Internet

• Very early start and a lot of experience
  • Pragmatic and evolutionary approach
  • If it's not broken, don't fix it
• Standardization by independent technical experts
  • Avoids the designed by committee effect of consortiums
  • Conservative and concentrating on stability
  • Implementations are required to prove technical feasibility
  • Simplicity whenever possible

(7) Internet Principles

Be liberal in what you accept, and conservative in what you send.

Jon Postel [http://www.postel.org/postel.html], IETF RFC 1122 [http://tools.ietf.org/html/rfc1122]

Whenever possible, communications protocol operations should be defined to occur at the end-points of a communications system, or as close as possible to the resource being controlled.

J. Saltzer, D. Reed, D. Clark, End-to-end Arguments in System Design [http://dret.net/biblio/reference/sal84]

(8) Internet Protocols

(9) Network Convergence

(11) Separation of Concerns (SoC)

• Solving different issues separately is a Good Idea [http://en.wikipedia.org/wiki/Separation_of_concerns#Origin]
  • ...a program must be correct...
  • ...it should be efficient...
  • ...whether, and if so: why, the program is desirable...
• Complex solutions need a structured approach for solving them
  • Solutions become easier to create
  • Parts of solutions may become reusable
  • Parts of solutions may become replaceable
• Software engineering hit complexity issues a bit earlier than networking did
• SoC is a good idea but not without costs
  • Vertical integration allows optimizations not possible in structured designs
  • Design depends largely on constraints. (Charles and Ray Eames [http://en.wikipedia.org/wiki/Charles_and_Ray_Eames])

(12) OSI Layers

(13) Layer 1: Physical

(14) Layer 2: Data Link

• Getting small units of data (frames) safely across layer 1
• Lost/damaged frames are repeated to ensure robustness
• Depending on layer 1, layer 2 frames can have very different sizes
  • Ethernet frames [http://en.wikipedia.org/wiki/Ethernet_frame] typically have 1500 bytes of payload (and some protocol fields)
  • Other networks (such as GSM, CDMA, BLE) use different frame sizes

(15) Layer 3: Network

• Allows addressing at the network level
• For the Internet, this means IP addresses
  • IPv4 [http://en.wikipedia.org/wiki/IPv4] is the version that has powered the Internet so far
  • IPv6 [http://en.wikipedia.org/wiki/IPv6] is phased in to be able to handle IoT-scale growth

(16) Layer 4: Transport

• Layer 3 establishes connectivity on a machine-to-machine level
  • Packets can be sent from one machine to another machine
  • It is not clear yet what to do with the packet once it arrived
• TCP establishes process-to-process connectivity and reliable connections

(17) Layer 5: Session

• Sessions are meaningful for applications
  • TCP connections may or may not terminate (see HTTP/1.1 persistent connections [http://en.wikipedia.org/wiki/HTTP_persistent_connection])
  • Applications should be able to start, pause, restart, and end sessions
• Sessions can be terrible for scalability
  • Creating tight coupling between two processes
  • REST [rest/] tries really hard to build a session-free web (stateless interactions)
• The Internet does not really have a generic model for sessions
  • HTTP Cookies are one way to achieve the same goal (but with a lot of side-effects)

(18) Layer 6: Presentation

• OSI has Abstract Syntax Notation One (ASN.1) [http://en.wikipedia.org/wiki/Abstract_Syntax_Notation_One]
  • one abstract model and a variety of encoding rules (DER, XER, PER)
  • mostly extinct nowadays but still used for PKI information (certificates) [http://en.wikipedia.org/wiki/Public_key_infrastructure]
• The Internet has media types and some generic data representation languages
  • Extensible Markup Language (XML) [../xml] was the first successful Internet representation
  • JavaScript Object Notation (JSON) replaced XML because most developers like it better []
  • Resource Description Framework (RDF) [] is another possible generic data format

(19) Layer 7: Application

• Anything that is built on top of a layered networking stack
• Applications can select (some of the) aspects of the stack
  • TCP is often useful because it's reliable and flow-controlled
  • UDP may be better for real-time and loss-tolerant data such as video/audio
• Lower layers are completely invisible to applications
  • Unless some support and signaling is built into the networking stack
  • For example, it can be very interesting to know what traffic costs
• Complete abstraction is a dream, complete control is a nightmare
  • Or is it the other way around? It all depends on the scenarios...
  • As the Internet and applications evolve, networking protocols and features evolve

(20) (Some) Popular Internet Application Protocols

• Domain/Host Name Resolution: DNS
• The Web: Hypertext Transfer Protocol (HTTP) [Representational State Transfer (REST) and Activity Streams; Hypertext Transfer Protocol (HTTP) (1)]
• Email: SMTP/IMAP
• Chatting: Extensible Messaging Protocol (XMPP) [IoT Communication Protocols; Extensible Messaging Protocol (XMPP) (1)]
• File Transfer: FTP
• Multimedia Traffic: RTP
• Remote Login: Telnet

(21) Popular Networking Equipment

• Layer 1: Repeaters, Hubs (extending reach)
• Layer 2: Bridges, Switches (selectively repeating frames)
• Layer 3: Routers (using end-to-end addresses)
• Above Layer 4: Gateways

(23) TCP Features

• Flow-controlled (avoiding congestion)
• Reliable (no data lost or duplicated)
• Connection-oriented
• Application addressing

(24) Reliable Connections

• IP may drop or duplicate packets
  • TCP adds serial numbers in data packets
  • If problems are detected, TCP recovers automatically
• TCP avoids network congestion and system overload
  • Slow start avoids flooding receivers with data they cannot process
  • Fast retransmit for avoiding timeouts when losing data
  • A Sliding window for controlling the amount of outstanding packets

(25) TCP Window

(27) Naming vs. Addressing

• IP addresses depend on network topology and organization
  • Reorganizing a network may change all IP addresses
  • Identifying important hosts should not be address-based
• Names are supposed to be more stable than addresses
  • A name is an abstract identification of something
  • Names can be used to obtain more information
• Network services should use names instead of addresses
  • Before using the service, a mapping has to be performed
  • The Domain Name System (DNS) is providing this service

(28) DNS Properties

• DNS has a bootstrap problem
  • DNS provides a service and should thus be identified by a name
  • For resolving names into addresses, the DNS service is required
• DNS configuration is part of basic Internet configuration
  • Dynamic Host Configuration Protocol (DHCP) provides IP address, netmask, gateway, and DNS server address
• DNS names are hierarchically structured
  • ischool.berkeley.edu, edu is the Top-Level Domain (TLD)
  • TLDs are either generic (gTLD) or country code (ccTLD)
  • Subdomains are federated (e.g., edu, us, uk, tv)

(29) Names Matter

• Names are not unique and namespaces are finite
  • Name disputes arise which were irrelevant before the Web
  • Cybersquatting as a popular way to make money
• Names can be worth a lot of money
  • business.com was sold for $7.5M in 1999 and again for $345M in 2007
• Name inflation can be used to generate money
  • aero, biz, coop, info, jobs, mobi, museum, name, pro, travel
  • Starting 2014, user-defined top-level domains went on sale [http://dret.typepad.com/dretblog/2008/06/dret.html]
• Names can have political significance
  • ccTLDs are assigned based on the UNO's idea of what a country is
• Names can have symbolic significance
  • Catalonia managed to get a domain of its own (cat)

(30) DNS Name Space

(31) DNS Namespace Organization

• Domain owners can organize the assignment of subdomains
  • berkeley.edu [http://www.berkeley.edu/] is an U.S. educational institution
  • ethz.ch [http://www.ethz.ch/] is a Swiss university
  • imperial.ac.uk [http://www.imperial.ac.uk/] is a British university
  • uts.edu.au [http://www.uts.edu.au/] is an Australian university
  • swiss.aero [http://www.swiss.aero/] makes money by selling new land
• Special rules may apply (Germany does not assign car license plate names)
• Organizations may be countries or companies
  • Countries have national organizations for assigning names
  • Companies may create an internal multi-level namespace (www.ischool.berkeley.edu [http://www.ischool.berkeley.edu/])

(32) Using DNS

• DNS is used by virtually all Internet applications
  • Names are more stable than addresses
• E-mail has some dedicated features built into DNS
  • Special entries (MX records) identify the e-mail server for a domain
  • Fallback entries help dealing with failing e-mail servers
• most URIs are based on DNS names
  • http://ischool.berkeley.edu/ identifies the access protocol and the host
  • The browser first performs a DNS lookup
  • A TCP connection is then established to the IP address returned by the DNS